feat: Used improved theme selections for tool call highlighting colors

- Hang fix: 120s timeout on compression LLM call + raw_stream
  channel-close handling (None => break instead of spinning)
- Supervisor effective active count: stop counting finished
  JoinHandles as occupying capacity slots
- Universal tool result size cap: truncate_if_needed() on
  ToolResult, applied in eval_tool_calls() after escalation block;
  configurable via max_tool_result_chars in AppConfig/AgentConfig
- Windowed compression: compression_keep_last config param keeps
  the N most recent messages visible after compression
- Fix pre-existing flaky test: add #[serial] to
  handle_list_available_unrestricted_when_no_whitelist so it does
  not race with TestConfigDirGuard-based tests that temporarily
  populate the agents data dir

.github/workflows/release.yaml

@@ -8,9 +8,9 @@ on:
   workflow_dispatch:
     inputs:
       bump_type:
-        description: "Specify the type of version bump"
+        description: 'Specify the type of version bump'
         required: true
-        default: "patch"
+        default: 'patch'
         type: choice
         options:
           - patch
@@ -46,7 +46,7 @@ jobs:
       - name: Set up Python
         uses: actions/setup-python@v4
         with:
-          python-version: "3.10"
+          python-version: '3.10'
 
       - name: Install Commitizen
         run: |
@@ -108,17 +108,19 @@ jobs:
 
           cargo update || true
 
+          sed -i "s|image: 'darkalex17/coyote:v[^']*'|image: 'darkalex17/coyote:v${VERSION}'|" assets/sbx-kit/spec.yaml
+
           # Git config that helps in Act
           git config user.name  "github-actions[bot]"
           git config user.email "github-actions[bot]@users.noreply.github.com"
           git config --global --add safe.directory "$GITHUB_WORKSPACE"
 
           git status --porcelain
-          git diff --name-only -- Cargo.toml Cargo.lock || true
+          git diff --name-only -- Cargo.toml Cargo.lock assets/sbx-kit/spec.yaml || true
 
-          if ! git diff --quiet -- Cargo.toml Cargo.lock; then
+          if ! git diff --quiet -- Cargo.toml Cargo.lock assets/sbx-kit/spec.yaml; then
-            git add -u -- Cargo.toml Cargo.lock
+            git add -u -- Cargo.toml Cargo.lock assets/sbx-kit/spec.yaml
-            git commit -m "chore: bump Cargo.toml to $VERSION"
+            git commit -m "chore: bump Cargo.toml and sandbox image to $VERSION"
           else
             echo "No changes to commit (already at $VERSION)"
           fi
@@ -163,28 +165,28 @@ jobs:
           - target: aarch64-unknown-linux-musl
             os: ubuntu-latest
             use-cross: true
-            cargo-flags: ""
+            cargo-flags: ''
           - target: aarch64-apple-darwin
             os: macos-latest
             use-cross: true
-            cargo-flags: ""
+            cargo-flags: ''
           - target: aarch64-pc-windows-msvc
             os: windows-latest
             use-cross: true
-            cargo-flags: ""
+            cargo-flags: ''
           - target: x86_64-apple-darwin
             os: macos-latest
-            cargo-flags: ""
+            cargo-flags: ''
           - target: x86_64-pc-windows-msvc
             os: windows-latest
-            cargo-flags: ""
+            cargo-flags: ''
           - target: x86_64-unknown-linux-musl
             os: ubuntu-latest
             use-cross: true
-            cargo-flags: ""
+            cargo-flags: ''
           - target: x86_64-unknown-linux-gnu
             os: ubuntu-latest
-            cargo-flags: ""
+            cargo-flags: ''
 
     steps:
       - name: Check if actor is repository owner
@@ -338,7 +340,7 @@ jobs:
             ${{ steps.package.outputs.archive }} 
             ${{ steps.package.outputs.sha }}
           tag_name: v${{ env.RELEASE_VERSION }}
-          name: "v${{ env.RELEASE_VERSION }}"
+          name: 'v${{ env.RELEASE_VERSION }}'
           body_path: artifacts/changelog.md
           prerelease: false
 
@@ -456,3 +458,63 @@ jobs:
         if: env.ACT != 'true'
         with:
           registry-token: ${{ secrets.CARGO_REGISTRY_TOKEN }}
+
+  publish-sandbox-image:
+    needs: [publish-github-release]
+    name: Publish Sandbox Docker Image
+    runs-on: ubuntu-latest
+    steps:
+      - name: Check if actor is repository owner
+        if: ${{ github.actor != github.repository_owner && env.ACT != 'true' }}
+        run: |
+          echo "You are not authorized to run this workflow."
+          exit 1
+
+      - name: Checkout repository
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 1
+
+      - name: Ensure repository is up-to-date
+        if: env.ACT != 'true'
+        run: |
+          git fetch --all
+          git pull
+
+      - name: Get release artifacts
+        uses: actions/download-artifact@v4
+        with:
+          path: artifacts
+          merge-multiple: true
+
+      - name: Set version variable
+        run: |
+          version="$(cat artifacts/release-version)"
+          echo "version=$version" >> $GITHUB_ENV
+
+      - name: Validate release environment variables
+        run: |
+          echo "Release version: ${{ env.version }}"
+
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Login to Docker Hub
+        if: env.ACT != 'true'
+        uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.DOCKER_USERNAME }}
+          password: ${{ secrets.DOCKER_PASSWORD }}
+
+      - name: Push to Docker Hub
+        uses: docker/build-push-action@v5
+        with:
+          context: .
+          file: Dockerfile
+          platforms: linux/amd64,linux/arm64
+          push: ${{ env.ACT != 'true' }}
+          tags: darkalex17/coyote:latest, darkalex17/coyote:${{ env.version }}
+          build-args: COYOTE_VERSION=${{ env.version }}

CHANGELOG.md

@@ -1,3 +1,163 @@
+## v0.7.4 (2026-07-02)
+
+### Feat
+
+- Pin specific usql version to sbx kit
+- recursively take ownership over the copied in coyote config for the sbx
+- explicitly specify the COYOTE_CONFIG_DIR in the sbx kit
+- --tail-logs can track log rollovers and incoporates a sleep timer to minimize idle CPU cycles
+- Added support for log rolling so log files don't just blow up over time
+
+### Fix
+
+- Added back in --kit specification for the running of the sbx
+- sbx isn't copying base files in their respective directories
+- Update deprecated sbx kit config
+- Properly chown the coyote config recursively and password file in the sbx
+
+## v0.7.3 (2026-06-24)
+
+### Fix
+
+- apply bootstrapping of functions at startup to fix edge case
+
+## v0.7.2 (2026-06-19)
+
+### Fix
+
+- usql version upgrade
+
+## v0.7.1 (2026-06-19)
+
+### Fix
+
+- sbx mixins must be passed in directories, not as files and the files must be named spec.yaml per new sbx version
+
+## v0.7.0 (2026-06-18)
+
+### Feat
+
+- added configurable cache path via the COYOTE_CACHE_PATH environment variable
+- added a memory option to .set tab completions
+- Added a diagnostic .info tools subcommand to make it easier to see what tools are enabled in all contexts
+- Added additional info outputs for enabled skills and sbx directories
+- directly execute shell commands from within the REPL
+- created mixin kit for built-in functions and MCP servers
+- Added sbx mixins for the secrets providers so users can also bootstrap those as well.
+- added support for loading sbx mixins that are dynamically discovered in the users workspace and config directory
+- Added a --fresh flag to let users create a truly bare bones sandbox without bootstrapping their config
+- initial built-in sandboxing support powered by Docker sbx
+- Added the ability to auto-bootstrap workspace memory when in git repos
+- Added explicit guardrail handling for pending agents
+- auto-append memory to memory index and don't necessarily require the LLM to remember to do it after a write
+- Added an --init-memory [global|workspace] flag to easily and quickly enable memory
+- added memory global configuration settings to the output of --info and .info
+- added .set memory REPL commands to control memory injection and applied formatting
+- Create the built-in memory management tools
+- Append the memory system prompts (readonly or r/w) to the system prompt when applicable
+- Created the --no-memory CLI flag to disable memory for this invocation
+- Added the memory configuration properties and storage to the main app config, roles, sessions, and agents.
+- initial scaffolding of a memory system
+
+### Fix
+
+- rebuild the tool scope after dynamically updating the skills_enabled value in the REPL
+- properly resolve Windows-based local vault password file locations and bootstrap them into the sandbox when possible
+- auto-translation of user-prefixed Mac and Linux paths for the vault password file when running inside a sandbox
+- don't attempt to auto complete .vault list in the REPL; that's the end of the command
+- buffer tool stdout as well as stderr so that any tools that error to stdout are captured and included in the response to the model, enabling the model to see what went wrong and to reason about how to fix it.
+- auto-bootstrapped memory was accidentally putting the MEMORY.md directly in the repo root rather than .coyote/memory/MEMORY.md
+- improved the fs_patch script description and added improved error handling to it.
+- added in forgotten require_max_tokens to the fable model
+- append memory functions to non-graph based agents on init
+- when auto_continue is disabled via the .set auto_continue false command, it should strip the todo functions from the list of functions
+- use rawPredict for non-streaming Claude requests
+
+### Refactor
+
+- Migrated the .skills command completion to use StateFlags and updated the help messages
+
+## v0.6.0 (2026-06-05)
+
+### Feat
+
+- added skill hint prompt injection and configuration
+- Fallthrough on missing secrets during mcp.json merging
+- validate visible_skills field at config load time
+- implemented reflexion (sorta) in sisyphus for significant code changes to delegate to the code-reviewer agent
+- improved explore agent
+- removed conditional fallback of LLM_*_RAW_JSON from built-ins
+- updated enabled_skills handling to support both list and comma-separated strings
+- added new REPL set commands for toggling skills and changing what skills are enabled
+- upgraded to the latest version of mcp-remote
+- fs_grep now works with both files and directories
+- improved code reviewer agents with skills
+- added round trip validation for vault providers to ensure permissions and authentication
+- created new first-time run wizard for secrets provider
+- vault_password_file or nothing at all is shorthand for just using the local gman provider for secret management
+- refactored gman usage to be generic and work with various vault providers and use the SupportedProvider enum directly for configurations
+- created initial parity gman generalization for vault provider
+- Refactored the sisyhpus agent system to utilize the new skills system to improve performance and reliability
+- llm graph nodes support skills
+- updated sisyphus and coder tools
+- removed potentially confusing tab completions for .skill
+- .edit skill <name> support from within the REPL
+- Added skills_dir to the info output of Coyote
+- Created a few auto built-in skills
+- Added support for auto_unload skills during chat
+- cleaned up skill implementation
+- support multiple skill flags to load multiple skills at CLI startup
+- Modified --skill CLI to allow users to specify skills to start the REPL or CLI with.
+- added CLI --skill flag for modifying skills easily
+- REPL integration with skills
+- dynamic loading/unloading of skill tools and MCP servers whenever load_skill/unload_skill are invoked
+- created built-in functions for listing, loading, and unloading skills
+- implemented the skills policy to track available skills per context
+- added remote install and install support for skills
+- created the skill registry
+- decided to make skills persist to disk like agents and not in-memory like built-in roles
+- scaffold skill module
+
+### Fix
+
+- disable skills for specific built-in roles
+- redirect stderr into user's /dev/tty for guards
+- azure doesn't support underscores in key vault
+- accidental regression on enabled_skills being empty = all
+- greedy secrets regex caused multiple secrets on one line to fail
+- add agent context check to skill visibility validation
+- enforced global visible_skills in llm node validation and improved skill loading error handling across the project
+- restore agent skill policy on error during effective policy calculation
+- apply the same validation for skill filenames on list_skills as happens everywhere else
+- the vault's init_bare should try to load the provisioned secret_provider from the config file without also interpolating any of the rest of the configuration file. It should only fail if the user has not yet created a configuration file; i.e. done a first-time run.
+- the vault roundtrip test used characters that are unsupported by some major secrets providers
+- fixed tool filtering logic for skills and user functions in agents
+- privilege leak when unloading skills and leaving tool scope untouched
+- When bootstrapping an app config to interpolate secrets, clone the secrets provider configuration as well so config secrets stored in remote vaults can be used properly
+- forgot to move back up the vault probe value error to be before the delete
+- don't silently fail on skill role composition extraction in llm nodes
+- set -euo pipefail for the temp script in execute_command.sh tool
+- added forgotten skill name validation to has_skill to prevent side-channel attacks
+- use unique values for the secrets round trip verification
+- stop interpolating a line if any errors occur
+- added path validation for skill names
+- effective_policy unconditionally overwrote skill values for role-like structs
+- updated execute_command to not mangle heredocs and also added explicit instructions to the coder and sisyphus agents to use fs_write and fs_patch over execute_command when writing files
+- llm nodes accidentally skipped skill_registry::effective_role because I was passing an inline role instead
+- updated temperature values for all agents and roles
+- added back in require_max_tokens for new Claude models
+- skill support also requires function calling to be enabled
+- non_tty tests break on some TTY terminals
+- skill loading on agents
+- forgot to bootstrap skills on REPL startup
+- remove now deprecated .skill edit command
+
+### Refactor
+
+- removed redundant skill name validation from has_skill function
+- support both CSV and list formats for enabled_tools
+- Support both CSV and list formats for enabled_mcp_servers
+
 ## v0.5.0 (2026-05-27)
 
 ### Feat

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "coyote-ai"
-version = "0.5.0"
+version = "0.7.4"
 edition = "2024"
 authors = ["Alex Clarke <alex.j.tusa@gmail.com>"]
 description = "An all-in-one, batteries included LLM CLI Tool"
@@ -17,6 +17,7 @@ exclude = [".github", "CONTRIBUTING.md"]
 anyhow = "1.0.69"
 bytes = "1.4.0"
 clap = { version = "4.5.40", features = ["cargo", "derive", "wrap_help"] }
+comfy-table = { version = "7.2.2", features = ["custom_styling"] }
 dirs = "6.0.0"
 dunce = "1.0.5"
 futures-util = "0.3.29"
@@ -49,7 +50,7 @@ textwrap = "0.16.0"
 ansi_colours = "1.2.2"
 eventsource-stream = "0.2.3"
 log = "0.4.28"
-log4rs = { version = "1.4.0", features = ["file_appender"] }
+log4rs = { version = "1.4.0", features = ["file_appender", "rolling_file_appender", "compound_policy", "fixed_window_roller", "size_trigger"] }
 shell-words = "1.1.0"
 sha2 = "0.10.8"
 unicode-width = "0.2.0"
@@ -58,6 +59,8 @@ http = "1.1.0"
 indexmap = { version = "2.2.6", features = ["serde"] }
 hmac = "0.12.1"
 aws-smithy-eventstream = "0.60.4"
+aws-smithy-types = "=1.4.9"
+time = "=0.3.47"
 urlencoding = "2.1.3"
 json-patch = { version = "4.0.0", default-features = false }
 bitflags = "2.5.0"
@@ -72,6 +75,7 @@ html_to_markdown = "0.1.0"
 rust-embed = "8.5.0"
 os_info = { version = "3.8.2", default-features = false }
 bm25 = { version = "2.0.1", features = ["parallelism"] }
+petgraph = { version = "0.7", features = ["serde-1"] }
 which = "8.0.0"
 fuzzy-matcher = "0.3.7"
 terminal-colorsaurus = "0.4.8"
@@ -104,6 +108,7 @@ self_update = { version = "0.44", default-features = false, features = [
   "archive-zip",
   "compression-zip-deflate",
 ] }
+qrcode = "0.14"
 
 [dependencies.reqwest]
 version = "0.13.3"

Dockerfile

@@ -0,0 +1,72 @@
+ARG COYOTE_VERSION
+FROM docker/sandbox-templates:shell-docker
+
+ARG COYOTE_VERSION
+ARG TARGETARCH
+
+ENV PATH="/home/agent/.cargo/bin:/home/agent/.local/bin:${PATH}"
+
+USER root
+
+RUN apt-get update && \
+    apt-get install -y --no-install-recommends \
+      jq curl git \
+      build-essential pkg-config \
+      cmake \
+      clang libclang-dev \
+      musl-tools \
+      libssl-dev \
+      pandoc \
+      bzip2 \
+      nano && \
+    rm -rf /var/lib/apt/lists/*
+
+RUN set -euo pipefail; \
+    USQL_VERSION=0.21.4; \
+    case "${TARGETARCH}" in \
+      amd64) USQL_ARCH=amd64 ;; \
+      arm64) USQL_ARCH=arm64 ;; \
+      *) echo "Unsupported TARGETARCH: ${TARGETARCH}" >&2; exit 1 ;; \
+    esac; \
+    TMPDIR=$(mktemp -d); \
+    curl -fsSL --retry 3 \
+      "https://github.com/xo/usql/releases/download/v${USQL_VERSION}/usql_static-${USQL_VERSION}-linux-${USQL_ARCH}.tar.bz2" \
+      -o "$TMPDIR/usql.tar.bz2"; \
+    tar -xjf "$TMPDIR/usql.tar.bz2" -C "$TMPDIR"; \
+    install -m 0755 "$TMPDIR/usql_static" /usr/local/bin/usql; \
+    rm -rf "$TMPDIR"
+
+USER 1000
+
+RUN curl -LsSf https://astral.sh/uv/install.sh | sh && \
+    printf '#!/bin/sh\nexec uv tool run "$@"\n' > "$HOME/.local/bin/uvx" && \
+    chmod +x "$HOME/.local/bin/uvx"
+
+RUN mkdir -p /usr/local/share/npm-global/lib
+
+RUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | \
+      sh -s -- -y --default-toolchain stable --profile minimal && \
+    . "$HOME/.cargo/env" && \
+    cargo install --locked iwec && \
+    cargo install --locked ast-grep
+
+USER root
+
+RUN set -euo pipefail; \
+    case "${TARGETARCH}" in \
+      amd64) MUSL_TARGET=x86_64-unknown-linux-musl ;; \
+      arm64) MUSL_TARGET=aarch64-unknown-linux-musl ;; \
+      *) echo "Unsupported TARGETARCH: ${TARGETARCH}" >&2; exit 1 ;; \
+    esac; \
+    TMPDIR=$(mktemp -d); \
+    curl -fsSL --retry 3 \
+      "https://github.com/Dark-Alex-17/coyote/releases/download/v${COYOTE_VERSION}/coyote-${MUSL_TARGET}.tar.gz" \
+      -o "$TMPDIR/coyote.tar.gz"; \
+    tar -xzf "$TMPDIR/coyote.tar.gz" -C "$TMPDIR"; \
+    install -m 0755 "$TMPDIR/coyote" /home/agent/.cargo/bin/coyote; \
+    chown 1000:1000 /home/agent/.cargo/bin/coyote; \
+    rm -rf "$TMPDIR"
+
+USER 1000
+
+ENTRYPOINT ["coyote"]

README.md

@@ -5,6 +5,7 @@
 ![Release](https://img.shields.io/github/v/release/Dark-Alex-17/coyote?color=%23c694ff)
 ![Crate.io downloads](https://img.shields.io/crates/d/coyote-ai?label=Crate%20downloads)
 [![GitHub Downloads](https://img.shields.io/github/downloads/Dark-Alex-17/coyote/total.svg?label=GitHub%20downloads)](https://github.com/Dark-Alex-17/coyote/releases)
+![Docker pulls](https://img.shields.io/docker/pulls/darkalex17/coyote?label=Docker%20downloads)
 
 Coyote is an all-in-one, batteries-included, LLM CLI tool featuring Shell Assistant, CLI & REPL Mode, RAG, AI Tools & 
 Agents, and More.
@@ -25,6 +26,7 @@ Coming from [AIChat](https://github.com/sigoden/aichat)? Follow the [migration g
 * [REPL](https://github.com/Dark-Alex-17/coyote/wiki/REPL): Interactive Read-Eval-Print Loop for conversational interactions with LLMs and Coyote.
   * [Custom REPL Prompt](https://github.com/Dark-Alex-17/coyote/wiki/REPL-Prompt): Customize the REPL prompt to provide useful contextual information.
 * [Vault](https://github.com/Dark-Alex-17/coyote/wiki/Vault): Securely store and manage sensitive information such as API keys and credentials.
+* [Sandboxes](https://github.com/Dark-Alex-17/coyote/wiki/Sandboxes): Launch Coyote inside an isolated [Docker Sandbox](https://docs.docker.com/ai/sandboxes/) with one command. Host config and vault credentials are projected in automatically; everything else is delegated to the `sbx` CLI.
 * [Shell Integrations](https://github.com/Dark-Alex-17/coyote/wiki/Shell-Integrations): Seamlessly integrate Coyote with your shell environment for enhanced command-line assistance.
 * [Function Calling](https://github.com/Dark-Alex-17/coyote/wiki/Tools): Leverage function calling capabilities to extend Coyote's functionality with custom tools
     * [Creating Custom Tools](https://github.com/Dark-Alex-17/coyote/wiki/Custom-Tools): You can create your own custom tools to enhance Coyote's capabilities.
@@ -36,6 +38,8 @@ Coming from [AIChat](https://github.com/sigoden/aichat)? Follow the [migration g
 * [Macros](https://github.com/Dark-Alex-17/coyote/wiki/Macros): Automate repetitive tasks and workflows with Coyote "scripts" (macros).
 * [RAG](https://github.com/Dark-Alex-17/coyote/wiki/RAG): Retrieval-Augmented Generation for enhanced information retrieval and generation.
 * [Sessions](https://github.com/Dark-Alex-17/coyote/wiki/Sessions): Manage and persist conversational contexts and settings across multiple interactions.
+* [Memory](https://github.com/Dark-Alex-17/coyote/wiki/Memory): Persistent file-based memory that survives across sessions. Bootstrap with `coyote --init-memory [global|workspace]`.
+* [Workspace Instructions](https://github.com/Dark-Alex-17/coyote/wiki/Workspace-Instructions): Human-curated project instructions (`COYOTE.md`) injected into every prompt, with `AGENTS.md`/`CLAUDE.md`/`GEMINI.md` fallbacks for cross-tool compatibility. Scaffold with `coyote --init-instructions`.
 * [Roles](https://github.com/Dark-Alex-17/coyote/wiki/Roles): Customize model behavior for specific tasks or domains.
 * [Skills](https://github.com/Dark-Alex-17/coyote/wiki/Skills): Modular knowledge or capability packs the LLM can load and unload mid-conversation. Multiple skills compose; instructions stack, tools and MCPs union.
 * [Agents](https://github.com/Dark-Alex-17/coyote/wiki/Agents): Leverage AI agents to perform complex tasks and workflows, including sub-agent spawning, teammate messaging, and user interaction tools.
@@ -57,6 +61,14 @@ Coyote requires the following tools to be installed on your system:
 * [docker](https://docs.docker.com/engine/install/)
 * [uv](https://docs.astral.sh/uv/getting-started/installation/)
     * `curl -LsSf https://astral.sh/uv/install.sh | sh`
+* [iwe](https://github.com/iwe-org/iwe) (`iwec`, for the built-in `iwe` MCP server that navigates large markdown knowledgebases)
+    * **Homebrew:** `brew tap iwe-org/iwe && brew trust --formula iwe-org/iwe/iwe && brew install iwe`
+    * **Cargo:** `cargo install iwec`
+* [ast-grep](https://ast-grep.github.io/) (for the built-in `ast_grep` structural code search tool, used by the `explore` agent)
+    * **Homebrew:** `brew install ast-grep`
+    * **Cargo:** `cargo install ast-grep --locked`
+    * **npm:** `npm i -g @ast-grep/cli`
+    * Optional: if `ast-grep` is not installed, the `ast_grep` tool reports it and agents fall back to `fs_grep`
 
 These tools are used to provide various functionalities within Coyote, such as document processing, JSON manipulation,
 etc., and they are used within agents and tools.
@@ -90,13 +102,39 @@ To upgrade `coyote` using Homebrew:
 brew upgrade coyote
 ```
 
+### Docker
+Coyote is available as a Docker image on Docker Hub (`darkalex17/coyote`) for Linux amd64 and arm64.
+Useful for CI, ephemeral environments, or anywhere you prefer not to install it natively.
+
+```bash
+docker pull darkalex17/coyote
+docker run --rm -it darkalex17/coyote
+```
+
+To persist your configuration across container runs, mount your existing config directory:
+
+```bash
+docker run --rm -it \
+  -v ~/.config/coyote:/home/agent/.config/coyote \
+  darkalex17/coyote
+```
+
+If you use the local vault provider and want your vault credentials available in the container, also mount the password file:
+
+```bash
+docker run --rm -it \
+  -v ~/.config/coyote:/home/agent/.config/coyote \
+  -v ~/.coyote_password:/home/agent/.coyote_password:ro \
+  darkalex17/coyote
+```
+
 ### Scripts
 #### Linux/MacOS (`bash`)
 You can use the following command to run a bash script that downloads and installs the latest version of `coyote` for your
 OS (Linux/MacOS) and architecture (x86_64/arm64):
 
 ```shell
-curl -fsSL https://raw.githubusercontent.com/Dark-Alex-17/coyote/main/install_coyote.sh | bash
+curl -fsSL https://raw.githubusercontent.com/Dark-Alex-17/coyote/refs/heads/main/scripts/install_coyote.sh | bash
 ```
 
 #### Windows/Linux/MacOS (`PowerShell`)
@@ -104,7 +142,7 @@ You can use the following command to run a PowerShell script that downloads and
 for your OS (Windows/Linux/MacOS) and architecture (x86_64/arm64):
 
 ```powershell
-powershell -NoProfile -ExecutionPolicy Bypass -Command "iwr -useb https://raw.githubusercontent.com/Dark-Alex-17/coyote/main/scripts/install_coyote.ps1 | iex"
+powershell -NoProfile -ExecutionPolicy Bypass -Command "iwr -useb https://raw.githubusercontent.com/Dark-Alex-17/coyote/refs/heads/main/scripts/install_coyote.ps1 | iex"
 ```
 
 ### Manual

assets/agents/adversary/README.md

@@ -0,0 +1,94 @@
+# Adversary
+
+An **adversarial plan-conformance reviewer**. Where [`code-reviewer`](../code-reviewer/README.md)
+asks *"is this code good?"*, `adversary` asks a different, harder question:
+
+> **"Is this the code the plan asked for — all of it, and only it?"**
+
+It hunts the gap between what a task/plan *specified* and what the implementer actually *built*:
+silently skipped acceptance criteria, scope creep, interface substitution, approach drift, and the
+requirements that never showed up in the diff at all ("the dog that didn't bark"). It assumes the
+implementer drifted until the diff proves otherwise — the independence is the value.
+
+## Why it's separate from `code-reviewer`
+
+| | `code-reviewer` | `adversary` |
+|---|---|---|
+| Question | Is the code correct/clean/safe? | Does the code match the plan? |
+| Input | The diff | The diff **+ the plan's acceptance criteria** |
+| Blind spot it covers | slop, bugs, coupling, footguns | skipped criteria, scope drift, contract breakage |
+| Output | severity-tagged findings (🔴🟡🟢) | a blocking verdict: `CONFORMS` / `DIVERGES` |
+
+They are **complementary passes**, not substitutes. `sisyphus` runs both on non-trivial work: one
+guards quality, the other guards fidelity to the plan.
+
+## Verdict (blocking)
+
+The agent ends every review with one sentinel:
+
+```
+ADVERSARIAL_REVIEW: CONFORMS
+Criteria: N/N met (all with tests).
+```
+
+```
+ADVERSARIAL_REVIEW: DIVERGES
+Criteria: X/N met, Y partial, Z unmet/diverged.
+Complaints:
+1. Acceptance criterion "<quoted>" — <Unmet|Partial|Diverged> — <what the diff does/omits, file:line> — <fix>
+2. ...
+```
+
+A `DIVERGES` verdict **blocks** completion. The caller (sisyphus/architect) must reconcile it —
+resume the SAME coder/sisyphus session with the complaints pasted verbatim — or escalate. It mirrors
+the `oracle` + `plan-review` gate used before implementation, but applied *after* implementation.
+
+Every complaint ties to a quoted acceptance criterion (or a named scope/interface/out-of-scope
+violation) and cites `file:line`. Vague complaints are not emitted.
+
+## How it reviews
+
+Driven by the [`adversarial-review`](../../skills/adversarial-review/SKILL.md) skill:
+
+1. Map **every** acceptance criterion to specific evidence in the diff → ✅ Met / ⚠️ Partial / ❌ Unmet / 🔀 Diverged. No test proving the behavior ⇒ at best ⚠️ Partial.
+2. Ground-truth with read-only tools (`fs_grep`/`fs_read`/`ast_grep`): confirm required symbols exist as specified, changes land where they must, new behavior is actually reached, tests target behavior not implementation.
+3. Hunt adversarially for the **absent**: skipped criteria, scope creep, interface/approach substitution, out-of-scope touches, downstream contract breakage.
+
+It is **read-only** — it produces a verdict, never a fix.
+
+## Usage
+
+Typically spawned by `sisyphus` (or `architect`) alongside `code-reviewer`. The spawn prompt IS its
+entire context, so it must include the diff (or a base ref to fetch) **and** the acceptance criteria:
+
+```sh
+agent__spawn --agent adversary --prompt "
+## TASK
+Adversarially review the recent changes for TASK-NNN against its plan. Return CONFORMS/DIVERGES.
+
+## DIFF
+Run get_diff (or --base main), or: <paste diff>
+
+## PLAN — acceptance criteria to check against
+<paste the task index.md body + the relevant PLAN-*.md section, verbatim>
+"
+```
+
+Direct invocation for ad-hoc use:
+
+```sh
+coyote -a adversary --agent-variable project_dir /path/to/repo \
+  "Review staged changes against these criteria: <paste criteria>"
+```
+
+### Tools
+
+- `get_diff [--base <ref>]` — staged → unstaged → `HEAD~1` fallback (or an explicit base/PR branch).
+- `get_changed_files [--base <ref>]` — quick changed-file map.
+- Plus read-only `fs_*` and `ast_grep` for ground-truth checks.
+
+## Related
+
+- [`adversarial-review`](../../skills/adversarial-review/SKILL.md) — the conformance methodology it runs on.
+- [`code-reviewer`](../code-reviewer/README.md) — the quality reviewer it runs alongside.
+- [`plan-review`](../../skills/plan-review/SKILL.md) — the *pre*-implementation plan gate; `adversary` is its *post*-implementation counterpart.

assets/agents/adversary/config.yaml

@@ -0,0 +1,118 @@
+name: adversary
+description: Adversarial plan-conformance reviewer - judges whether an implementation matches the task/plan it was supposed to satisfy (not code quality). Returns a blocking CONFORMS/DIVERGES verdict. Complements code-reviewer. Designed to be delegated to by sisyphus.
+version: 1.0.0
+
+auto_continue: true
+max_auto_continues: 15
+inject_todo_instructions: true
+
+skills_enabled: true
+enabled_skills:
+  - adversarial-review
+
+variables:
+  - name: project_dir
+    description: Project directory containing the changes under review
+    default: '.'
+  - name: auto_confirm
+    description: Auto-confirm command execution
+    default: '1'
+
+global_tools:
+  - ast_grep.sh
+  - fs_read.sh
+  - fs_cat.sh
+  - fs_grep.sh
+  - fs_glob.sh
+  - fs_ls.sh
+  - execute_command.sh
+
+instructions: |
+  You are an adversarial plan-conformance reviewer. You answer ONE question: **does this
+  implementation match the plan it was supposed to satisfy — all of it, and only it?** You are NOT
+  the code-quality reviewer (that is `code-reviewer`/`file-reviewer`, which judges correctness, slop,
+  and style). You judge CONFORMANCE: skipped acceptance criteria, silent scope drift, interface
+  substitution, and things the plan required that never showed up in the diff.
+
+  Your value is independence and suspicion. Assume the implementer drifted, cut a corner, or misread
+  the plan until the diff proves otherwise.
+
+  ## Step 0: Load the skill
+
+  Before anything else, `skill__load` `adversarial-review`. It carries your methodology: the
+  criterion-by-criterion evidence mapping, the adversarial checklist (silently skipped criteria,
+  scope drift, interface drift, ground-truth verification, out-of-scope violations, downstream
+  contract breakage), and the exact verdict format. The skill body is your source of truth for HOW to
+  review and WHAT to flag; these instructions handle workflow and I/O.
+
+  ## Input (the spawn prompt IS your entire context)
+
+  You are given:
+  1. **The diff** — pasted inline, or run `get_diff` (optionally `--base <ref>`) if told to fetch it.
+  2. **The plan** — the task's Objective, Tasks, and especially its **Acceptance criteria**, pasted
+     inline (e.g. a BCP task `index.md` body + the relevant `PLAN-*.md` section), or a path to read.
+
+  If the plan / acceptance criteria are missing, STOP and say so: conformance cannot be judged
+  without a spec. Do not invent criteria or guess intent.
+
+  ## Workflow
+
+  1. Load `adversarial-review`.
+  2. Get the diff (inline or via `get_diff`) and identify the changed files.
+  3. For EACH acceptance criterion: find the specific evidence in the diff that satisfies it and
+     classify it ✅ Met / ⚠️ Partial / ❌ Unmet / 🔀 Diverged. A criterion with no test proving its
+     behavior is at best ⚠️ Partial.
+  4. Ground-truth every claim: `fs_grep` the symbols the plan requires (confirm they exist, spelled
+     as specified), `fs_read` around each hunk to confirm the change makes the criterion true, grep
+     callers to confirm new behavior is reached, confirm tests target behavior not implementation.
+     Use `ast_grep` for structural checks (e.g. "was this function signature actually changed?").
+  5. Hunt adversarially for what's ABSENT (the dog that didn't bark), scope creep, interface/approach
+     substitution, out-of-scope touches, and downstream contract breakage — per the skill checklist.
+  6. Emit the verdict in the skill's exact format.
+
+  ## Output — verdict (MANDATORY, exact format)
+
+  End with EXACTLY one of these sentinels so the caller can route on it:
+
+  ```
+  ADVERSARIAL_REVIEW: CONFORMS
+  Criteria: N/N met (all with tests).
+  <optional: 1-3 non-blocking observations>
+  ```
+
+  ```
+  ADVERSARIAL_REVIEW: DIVERGES
+  Criteria: X/N met, Y partial, Z unmet/diverged.
+  Complaints:
+  1. Acceptance criterion "<quoted>" — <Unmet|Partial|Diverged> — <what the diff does/omits, file:line> — <what would make it conform>
+  2. Scope drift / interface drift / out-of-scope — <file:line> — <the violation> — <the fix>
+  3. ...
+  ```
+
+  Every complaint MUST quote the specific acceptance criterion (or name the specific scope/interface/
+  out-of-scope violation) AND cite file:line. A complaint with no criterion reference and no location
+  is noise — do not emit it.
+
+  ## Rules
+
+  1. **You are read-only.** Never modify files. You produce a verdict; the implementer owns the fix.
+  2. **Conformance, not quality.** Do not flag style/naming/micro-optimizations unless they cause a
+     criterion to be unmet. If a quality defect breaks a criterion (a race violating a correctness
+     criterion), flag it as a conformance failure and note it is also a quality issue.
+  3. **No test ⇒ not met.** An acceptance criterion is a promise of observable behavior; unproven
+     behavior is at best Partial.
+  4. **Absence is a finding.** Review what SHOULD be in the diff per the plan, not only what IS.
+  5. **Don't re-litigate a settled decision** — but DO flag when the diff silently overrode one the
+     plan recorded ("do X not Y because Z" → diff does Y).
+  6. **The plan can be the culprit.** If the plan is impossible/self-contradictory, that is DIVERGES
+     with the plan named as root cause — never judge against a plan you silently corrected.
+  7. Be terse and decisive. Three real divergences beat fifteen weak ones. If everything is a nitpick,
+     it CONFORMS — say so.
+
+  ## Context
+  - Project: {{project_dir}}
+  - CWD: {{__cwd__}}
+  - Shell: {{__shell__}}
+
+  ## Available Tools
+  {{__tools__}}

assets/agents/adversary/tools.sh

@@ -0,0 +1,78 @@
+#!/usr/bin/env bash
+set -eo pipefail
+
+# @env LLM_OUTPUT=/dev/stdout
+# @env LLM_AGENT_VAR_PROJECT_DIR=.
+# @describe Adversarial plan-conformance reviewer tools
+
+_project_dir() {
+  local dir="${LLM_AGENT_VAR_PROJECT_DIR:-.}"
+  (cd "${dir}" 2>/dev/null && pwd) || echo "${dir}"
+}
+
+# @cmd Get the git diff to review for plan conformance. Returns staged changes, or unstaged if nothing is staged, or the HEAD~1 diff if the working tree is clean.
+# @option --base Optional base ref to diff against (e.g., "main", "HEAD~3", a commit SHA, or a PR base branch)
+get_diff() {
+  local project_dir
+  project_dir=$(_project_dir)
+  # shellcheck disable=SC2154
+  local base="${argc_base:-}"
+
+  local diff_output=""
+  if [[ -n "${base}" ]]; then
+    diff_output=$(cd "${project_dir}" && git diff "${base}" 2>&1) || true
+  else
+    diff_output=$(cd "${project_dir}" && git diff --cached 2>&1) || true
+    if [[ -z "${diff_output}" ]]; then
+      diff_output=$(cd "${project_dir}" && git diff 2>&1) || true
+    fi
+    if [[ -z "${diff_output}" ]]; then
+      diff_output=$(cd "${project_dir}" && git diff HEAD~1 2>&1) || true
+    fi
+  fi
+
+  if [[ -z "${diff_output}" ]]; then
+    echo "No changes found to review in ${project_dir}." >> "$LLM_OUTPUT"
+    return 0
+  fi
+
+  local file_count
+  file_count=$(echo "${diff_output}" | grep -c '^diff --git' || true)
+  {
+    echo "Diff contains changes to ${file_count} file(s):"
+    echo ""
+    echo "${diff_output}"
+  } >> "$LLM_OUTPUT"
+}
+
+# @cmd Get the list of changed files with stats (a quick map of what to check against the plan).
+# @option --base Optional base ref to diff against
+get_changed_files() {
+  local project_dir
+  project_dir=$(_project_dir)
+  local base="${argc_base:-}"
+
+  local stat_output=""
+  if [[ -n "${base}" ]]; then
+    stat_output=$(cd "${project_dir}" && git diff --stat "${base}" 2>&1) || true
+  else
+    stat_output=$(cd "${project_dir}" && git diff --cached --stat 2>&1) || true
+    if [[ -z "${stat_output}" ]]; then
+      stat_output=$(cd "${project_dir}" && git diff --stat 2>&1) || true
+    fi
+    if [[ -z "${stat_output}" ]]; then
+      stat_output=$(cd "${project_dir}" && git diff --stat HEAD~1 2>&1) || true
+    fi
+  fi
+
+  if [[ -z "${stat_output}" ]]; then
+    echo "No changes found in ${project_dir}." >> "$LLM_OUTPUT"
+    return 0
+  fi
+
+  {
+    echo "Changed files:"
+    echo ""
+    echo "${stat_output}"
+  } >> "$LLM_OUTPUT"
+}

assets/agents/code-reviewer/README.md

@@ -16,7 +16,7 @@ agents while handling coordination and final reporting.
 ## Pro-Tip: Use an IDE MCP Server for Improved Performance
 Many modern IDEs now include MCP servers that let LLMs perform operations within the IDE itself and use IDE tools. Using
 an IDE's MCP server dramatically improves the performance of coding agents. So if you have an IDE, try adding that MCP
-server to your config (see the [MCP Server docs](../../../docs/function-calling/MCP-SERVERS.md) to see how to configure
+server to your config (see the [MCP Server docs](https://github.com/Dark-Alex-17/coyote/wiki/MCP-Servers) to see how to configure
 them), and modify the agent definition to look like this:
 
 ```yaml

assets/agents/code-reviewer/config.yaml

@@ -19,8 +19,12 @@ variables:
   - name: project_dir
     description: Project directory to review
     default: '.'
+  - name: auto_confirm
+    description: Auto-confirm command execution
+    default: '1'
 
 global_tools:
+  - ast_grep.sh
   - fs_read.sh
   - fs_cat.sh
   - fs_grep.sh

assets/agents/coder/graph.yaml

@@ -2,9 +2,9 @@ name: coder
 description: |
   Implementation agent. Plans, implements, and runs build + tests in a
   bounded fix-loop until verified. Designed to be delegated to by sisyphus.
-version: "1.0"
+version: '1.0'
-
 global_tools:
+  - ast_grep.sh
   - fs_cat.sh
   - fs_ls.sh
   - fs_write.sh
@@ -25,7 +25,7 @@ variables:
       Absolute path to the project directory. Defaults to "." which is the
       directory you invoked `coyote` from. Override at runtime with
       `coyote -a coder --agent-variable project_dir /abs/path "..."`.
-    default: "."
+    default: '.'
 
 settings:
   max_loop_iterations: 20
@@ -34,14 +34,14 @@ settings:
   timeout: 1800
 
 initial_state:
-  project_dir: ""
+  project_dir: ''
   fix_attempts: 0
   max_fix_attempts: 3
-  fix_instructions: ""
+  fix_instructions: ''
-  build_output: ""
+  build_output: ''
-  tests_output: ""
+  tests_output: ''
-  last_node_output: ""
+  last_node_output: ''
-  plan_summary: ""
+  plan_summary: ''
   files_to_modify: []
   files_to_create: []
   risks: []
@@ -49,7 +49,7 @@ initial_state:
   review_attempts: 0
   max_review_attempts: 1
   review_clean: true
-  review_notes: ""
+  review_notes: ''
 
 start: resolve_paths
 
@@ -88,7 +88,7 @@ nodes:
       etc. Empty list is fine.
 
       Project directory: {{project_dir}}
-    prompt: "{{initial_prompt}}"
+    prompt: '{{initial_prompt}}'
     tools: []
     output_schema:
       type: object
@@ -98,20 +98,27 @@ nodes:
           description: 1-3 sentences summarizing what will be done
         files_to_modify:
           type: array
-          items: {type: string}
+          items: { type: string }
         files_to_create:
           type: array
-          items: {type: string}
+          items: { type: string }
         complexity_score:
           type: integer
           minimum: 1
           maximum: 10
         risks:
           type: array
-          items: {type: string}
+          items: { type: string }
-      required: [plan_summary, files_to_modify, files_to_create, complexity_score, risks]
+      required:
+        [
+          plan_summary,
+          files_to_modify,
+          files_to_create,
+          complexity_score,
+          risks,
+        ]
     state_updates:
-      last_node_output: "{{output}}"
+      last_node_output: '{{output}}'
     fallback: end_failure
     next: route_complexity
 
@@ -144,11 +151,11 @@ nodes:
 
       Approve this plan?
     options:
-      - "yes"
+      - 'yes'
-      - "no"
+      - 'no'
     routes:
-      "yes": implement
+      'yes': implement
-      "no": end_rejected
+      'no': end_rejected
     on_other: end_rejected
 
   implement:
@@ -243,7 +250,7 @@ nodes:
       - execute_command
     max_iterations: 30
     state_updates:
-      last_node_output: "{{output}}"
+      last_node_output: '{{output}}'
     fallback: end_failure
     next: verify_build
 
@@ -326,7 +333,7 @@ nodes:
           description: Concrete issues found, one per line as file:line - description. Empty when review_clean is true.
       required: [review_clean, review_notes]
     state_updates:
-      last_node_output: "{{output}}"
+      last_node_output: '{{output}}'
     fallback: end_success
     next: route_review_result
 

assets/agents/explore/config.yaml

@@ -1,6 +1,6 @@
 name: explore
-description: Fast codebase exploration agent - finds patterns, structures, and relevant files. Designed to be fanned out 2-5 in parallel by orchestrators.
+description: Fast codebase exploration agent - finds patterns, structures, and relevant files. Designed to be fanned out in parallel by orchestrators — scale to the number of distinct search angles the task requires.
-version: 3.0.0
+version: 3.1.0
 
 skills_enabled: true
 enabled_skills:
@@ -10,10 +10,14 @@ variables:
   - name: project_dir
     description: Project directory to explore
     default: '.'
+  - name: auto_confirm
+    description: Auto-confirm command execution
+    default: '1'
 
 mcp_servers:
   - ddg-search
 global_tools:
+  - ast_grep.sh
   - fs_read.sh
   - fs_cat.sh
   - fs_grep.sh
@@ -33,7 +37,7 @@ instructions: |
 
   ## You may be one of many parallel explorers
 
-  Orchestrators (like Sisyphus) often fan out 2-5 explore agents at once, each covering a different angle of the same question. Assume you are ONE narrow slice of a larger investigation. Stay strictly within YOUR slice as defined by the prompt — don't broaden scope to cover what other parallel explorers might be handling.
+  Orchestrators (like Sisyphus) fan out as many explore agents as the task warrants — one per distinct search angle, module boundary, or concern. You may be one of many running in parallel. Assume you are ONE narrow slice of a larger investigation. Stay strictly within YOUR slice as defined by the prompt — don't broaden scope to cover what other parallel explorers might be handling.
 
   If the prompt says "find auth middleware", you find auth middleware. You do NOT also tour the routing layer, the error system, and the database connection pool. Narrow scope is the contract.
 
@@ -49,6 +53,8 @@ instructions: |
 
   4. **Locate symbols with `fs_grep`** — for finding where things live across the codebase. `fs_grep --pattern "fn handle_request" --include "*.rs"` is faster than reading files.
 
+  4b. **Match code STRUCTURE with `ast_grep`** — when text grep is too noisy or formatting-dependent. It matches syntax trees: `ast_grep --pattern '$X.unwrap()' --lang rust` finds every unwrap call however it's formatted; `ast_grep --pattern 'fn $NAME($$$) { $$$ }' --lang rust --glob 'src/**'` finds function definitions; `ast_grep --pattern 'useEffect($$$)' --lang tsx` finds hook usages that a text grep for "useEffect" would bury in comments and strings. Meta-variables: `$NAME` = one AST node, `$$$` = zero or more. The pattern must be a COMPLETE, valid AST node for `--lang` — `fn $NAME($$$)` without a body parses as nothing and matches nothing. Use `fs_grep` for plain text, comments, strings, and config files; `ast_grep` for calls, definitions, and signatures. If ast-grep isn't installed the tool says so — fall back to fs_grep.
+
   5. **Read targeted sections with `fs_read --offset/--limit`** — `fs_read --path "src/main.rs" --offset 50 --limit 30` reads lines 50-79 only. `fs_read` adds line numbers but TRUNCATES long lines (over 2000 chars) and caps output at 2000 lines by default.
 
   6. **Use `fs_cat` only when you need the full untruncated file** — rare in exploration. If you reach for `fs_cat`, ask whether `fs_grep` + targeted `fs_read` would answer your question with less context spend.
@@ -59,6 +65,7 @@ instructions: |
 
   - `fs_grep --pattern "struct User" --include "*.rs"` — find content across files in a directory tree
   - `fs_grep --pattern "TODO" --path "src/main.rs"` — find content within a single file (--include is ignored in this mode)
+  - `ast_grep --pattern 'impl $TRAIT for $TYPE' --lang rust` — find code by STRUCTURE, not text (see 4b above)
   - `fs_glob --pattern "*.rs" --path src/` — find files by name pattern
   - `fs_read --path "src/main.rs"` — read a TRUNCATED view with line numbers (default 2000 lines, lines over 2000 chars cut off)
   - `fs_read --path "src/main.rs" --offset 100 --limit 50` — read lines 100-149 only (line numbers; truncation rules still apply)

assets/agents/file-reviewer/README.md

@@ -16,7 +16,7 @@ one file while communicating with sibling agents to catch issues that span multi
 ## Pro-Tip: Use an IDE MCP Server for Improved Performance
 Many modern IDEs now include MCP servers that let LLMs perform operations within the IDE itself and use IDE tools. Using
 an IDE's MCP server dramatically improves the performance of coding agents. So if you have an IDE, try adding that MCP
-server to your config (see the [MCP Server docs](../../../docs/function-calling/MCP-SERVERS.md) to see how to configure
+server to your config (see the [MCP Server docs](https://github.com/Dark-Alex-17/coyote/wiki/MCP-Servers) to see how to configure
 them), and modify the agent definition to look like this:
 
 ```yaml

assets/agents/file-reviewer/config.yaml

@@ -11,6 +11,9 @@ variables:
   - name: project_dir
     description: Project directory for context
     default: '.'
+  - name: auto_confirm
+    description: Auto-confirm command execution
+    default: '1'
 
 global_tools:
   - fs_read.sh

assets/agents/librarian/graph.yaml

@@ -8,8 +8,6 @@ description: |
   sisyphus alongside explore when unfamiliar libraries/APIs/frameworks are
   involved.
 
-  Iteration 3: smart triage node up front + final-format trim of LLM
-  narrative leakage.
 version: "1.0"
 
 global_tools:

assets/agents/oracle/config.yaml

@@ -1,20 +1,27 @@
 name: oracle
 description: High-IQ advisor for architecture, debugging, and complex decisions. Blocking by design - the orchestrator is waiting on you.
-version: 2.0.0
+version: 2.1.0
 
 skills_enabled: true
 enabled_skills:
   - code-review
   - ai-slop-remover
+  - plan-review
+  - plan-authoring
+  - iwe-knowledge-base
 
 variables:
   - name: project_dir
     description: Project directory for context
     default: '.'
+  - name: auto_confirm
+    description: Auto-confirm command execution
+    default: '1'
 
 mcp_servers:
   - ddg-search
 global_tools:
+  - ast_grep.sh
   - fs_read.sh
   - fs_cat.sh
   - fs_grep.sh
@@ -46,13 +53,16 @@ instructions: |
   3. **Code review** — evaluating proposed designs or implementations.
   4. **Risk assessment** — security, performance, reliability concerns.
   5. **Multi-component questions** — anything spanning 3+ files or modules.
+  6. **Plan review** — critiquing implementation plans (high-level or per-step) BEFORE execution begins.
 
   ## Skills available
 
-  Two skills are available to you. Load them when relevant:
+  Load skills when relevant:
 
   - `skill__load code-review` — when reviewing a diff or existing code; gives you a focused review checklist.
   - `skill__load ai-slop-remover` — when judging code quality (especially for advising on cleanups).
+  - `skill__load plan-review` — when asked to review an implementation plan; adversarial checklist plus the PLAN_REVIEW verdict format. Load `plan-authoring` alongside it — it defines the plan schema you are checking against.
+  - `skill__load iwe-knowledge-base` — when the plans live in a large markdown corpus; navigate it structurally instead of globbing.
 
   Use `skill__list` to see what's available; `skill__unload` when done to keep context lean.
 
@@ -91,6 +101,8 @@ instructions: |
   ORACLE_COMPLETE
   ```
 
+  Exception: for plan reviews, use the `PLAN_REVIEW: OKAY` / `PLAN_REVIEW: REJECT` verdict format from the `plan-review` skill as the body, then end with `ORACLE_COMPLETE` on the final line as usual.
+
   ## Rules
 
   1. **Never modify files** — you advise, others implement.

assets/agents/sisyphus/README.md

@@ -16,6 +16,21 @@ Sisyphus acts as the primary entry point, capable of handling complex tasks by c
 - 💻 **CLI Coding**: Provides a natural language interface for writing and editing code.
 - 🔄 **Task Management**: Tracks progress and context across complex operations.
 - 🛠️ **Tool Integration**: Seamlessly uses system tools for building, testing, and file manipulation.
+- 📋 **Plan-Driven Workflows**: Authors, reviews, and executes phased implementation plans with handoffs between steps.
+
+## Plan-Driven Workflows
+
+For large features, Sisyphus supports a phased workflow backed by a plan repo (`plans/` with `steps/`, `handoffs/`, and
+a rolling `NOTES.md`):
+
+1. **Author** — after converging on a solution with you, Sisyphus loads the `plan-authoring` skill and writes a
+   high-level plan plus one grounded, self-contained implementation plan per step.
+2. **Review** — [Oracle](../oracle/README.md) critiques the plans with the `plan-review` skill (ground-truth checks
+   against the codebase, verifiability, dependency ordering) and returns a `PLAN_REVIEW: OKAY`/`REJECT` verdict.
+   Rejected plans are fixed before any code is written.
+3. **Execute** — one step at a time via the `step-implementation` and `handoff-protocol` skills: read the previous
+   handoff, staleness-check the plan, implement (delegating to [Coder](../coder/README.md)), verify, review, write an
+   evidence-backed handoff, and stop for your approval before the next step begins.
 
 ## Pro-Tip: Use an IDE MCP Server for Improved Performance
 Many modern IDEs (JetBrains, VS Code, Cursor, Zed, etc.) expose MCP servers that let LLMs use IDE tools directly. Using

assets/agents/sisyphus/config.yaml

@@ -1,6 +1,6 @@
 name: sisyphus
 description: OpenCode-style orchestrator - classifies intent, delegates to specialists, tracks progress with todos, enforces OMO-grade verification discipline
-version: 3.0.0
+version: 3.2.0
 
 agent_session: temp
 auto_continue: true
@@ -8,6 +8,14 @@ max_auto_continues: 25
 inject_todo_instructions: true
 
 can_spawn_agents: true
+spawnable_agents:
+  - explore
+  - librarian
+  - coder
+  - oracle
+  - code-reviewer
+  - adversary
+  - step-runner
 max_concurrent_agents: 4
 max_agent_depth: 3
 inject_spawn_instructions: true
@@ -23,6 +31,10 @@ enabled_skills:
   - parallel-research
   - verification-gates
   - oracle-protocol
+  - plan-authoring
+  - step-implementation
+  - handoff-protocol
+  - iwe-knowledge-base
 
 variables:
   - name: project_dir
@@ -35,6 +47,7 @@ variables:
 mcp_servers:
   - ddg-search
 global_tools:
+  - ast_grep.sh
   - fs_read.sh
   - fs_grep.sh
   - fs_glob.sh
@@ -101,6 +114,9 @@ instructions: |
   | About to touch git history | `git-master` |
   | About to touch UI/components | `frontend-ui-ux` (also nudge delegates to load it) |
   | About to write any code | `ai-slop-remover` |
+  | About to author a high-level plan or step plans | `plan-authoring` |
+  | About to execute a step of a phased plan | `step-implementation` + `handoff-protocol` |
+  | Navigating a plan repo or markdown knowledge base | `iwe-knowledge-base` |
 
   Load skills BEFORE the phase, not after. Unload when the phase ends if context is getting heavy. `skill__unload` keeps the context lean.
 
@@ -121,10 +137,11 @@ instructions: |
 
   | Agent | Use For | Characteristics |
   |-------|---------|-----------------|
-  | `explore` | Find patterns in THIS codebase, understand local code | Read-only, returns findings, fan out 2-5 in parallel |
+  | `explore` | Find patterns in THIS codebase, understand local code | Read-only, returns findings, fan out as many as the task warrants — one per distinct search angle, module, or concern. Large codebases or cross-cutting tasks should spawn 5–15+. |
-  | `librarian` | Find official docs, OSS examples, web best practices for EXTERNAL libraries | Read-only, returns citation-backed findings, fan out 1-3 in parallel |
+  | `librarian` | Find official docs, OSS examples, web best practices for EXTERNAL libraries | Read-only, returns citation-backed findings, fan out as many as distinct external sources or questions warrant — typically 2–6, more if the topic spans multiple libraries or specs. |
   | `coder` | Write/edit files, implement features | Graph agent: plan → approval → implement → verify build+tests → self_review → bounded fix-loop |
-  | `oracle` | Architecture, complex debugging, review | Advisory, blocking — never answer the user before collecting Oracle results |
+  | `oracle` | Architecture, complex debugging, review, plan review | Advisory, blocking — never answer the user before collecting Oracle results |
+  | `step-runner` | Execute ONE step of a phased plan repo (Phase 8) | Graph agent: orient → staleness check → coder → verify → handoff → user approval gate |
 
   ### When to fire `librarian` (external grep) vs `explore` (internal grep)
 
@@ -186,7 +203,17 @@ instructions: |
 
   ## Phase 4 - Parallel Research
 
-  When delegating exploration, load `parallel-research` skill, then fan out 2-5 `explore` agents in parallel, each scoped to a different angle. Each gets a NARROW slice.
+  When delegating exploration, load `parallel-research` skill, then fan out `explore` agents in parallel — one per distinct search angle, module boundary, or concern. Each gets a NARROW slice. Scale to the task:
+
+  | Task scope | Suggested fan-out |
+  |---|---|
+  | Single feature, known location | 2–3 |
+  | Multi-file feature across 2-3 modules | 4–6 |
+  | Cross-cutting concern (auth, error handling, config) across whole codebase | 7–12 |
+  | Large refactor or architectural analysis spanning many modules | 10–20+ |
+  | Full codebase audit (security, performance, pattern consistency) | One agent per top-level module or package |
+
+  Never artificially cap at a small number. If there are 10 distinct things to find, spawn 10 agents. The system limit is the only ceiling that matters.
 
   ### The wait protocol
 
@@ -278,6 +305,31 @@ instructions: |
 
   After a fix-loop completes, do not automatically re-run `code-reviewer` unless the fix itself triggers the same thresholds (2+ coders, 5+ files, architectural). Each `code-reviewer` invocation fans out N file-reviewers per changed file; spurious re-runs burn budget without proportional value. Trust coder's `self_review` on bounded fixes.
 
+  ### Adversarial plan-conformance review (post-coder, when the work implements a plan/spec)
+
+  `code-reviewer` asks "is this code good?" It does NOT check "is this the code the plan asked for?" When the coder work implemented against a written spec — a task file, a `plans/` step, an acceptance-criteria list, or any request with explicit "done when …" criteria — spawn `adversary` for an independent conformance pass. It maps every acceptance criterion to evidence in the diff and hunts for silently-skipped criteria, scope drift, interface substitution, and requirements that never landed ("the dog that didn't bark").
+
+  **When to spawn it:** whenever the change has a checkable spec. This is orthogonal to the `code-reviewer` thresholds — a one-file change can still silently skip an acceptance criterion. If there is a plan/task/criteria list, run `adversary`. Run BOTH reviewers when the work is both broad (code-reviewer thresholds fire) AND spec-driven; they cover different failure modes and their prompts differ (code-reviewer gets the diff; adversary gets the diff PLUS the acceptance criteria).
+
+  **Spawn pattern** (the prompt IS its whole context — it MUST include the criteria):
+
+  ```
+  agent__spawn --agent adversary --prompt "Adversarially review the recent coder change(s) for conformance to the plan. Return CONFORMS/DIVERGES.
+
+  DIFF: run get_diff (or --base <ref>), or: <paste diff>
+
+  PLAN — acceptance criteria to check against:
+  <paste the task/step spec + acceptance criteria VERBATIM — not a summary>"
+  ```
+
+  ### Handling adversary findings
+
+  - **`ADVERSARIAL_REVIEW: DIVERGES` blocks completion.** Do not mark the task done. Resume the SAME coder session (`agent__spawn --session_id <id> --prompt "Fix these plan-conformance failures: <complaints pasted verbatim>"`) — do not spawn a fresh coder. After the fix, re-run `adversary` ONCE to confirm it now CONFORMS; if it still DIVERGES on the same criteria after one fix cycle, STOP and escalate to the user (the plan or the approach may be wrong — consider `oracle`).
+  - **`ADVERSARIAL_REVIEW: CONFORMS`** — conformance satisfied; proceed (subject to code-reviewer's quality findings still being resolved).
+  - **A complaint that the PLAN itself is the root cause** (impossible/contradictory criterion) — do NOT silently "fix" by changing scope. Surface it to the user; the plan needs amending, which is their call.
+
+  Unlike `code-reviewer`, re-running `adversary` once after a conformance fix is expected — a DIVERGES verdict is a hard gate, and confirming the fix actually closed it is the point.
+
   ## File Operations (Direct Edits)
 
   When you write or modify files yourself (rather than delegating to coder):
@@ -312,6 +364,47 @@ instructions: |
 
   Never: leave code in broken state, continue hoping it'll work, delete failing tests to "pass," suppress errors to silence them.
 
+  ## Phase 8 - Plan-Driven Work (phased implementation via a plan repo)
+
+  Detect this mode when the user references step plans, handoffs, or a plan repo — or the workspace contains `plans/` with `steps/` and `handoffs/`. Plan-driven work has two lifecycles. Never mix them in one turn.
+
+  ### Authoring lifecycle (no code changes)
+
+  1. Discuss the problem; converge on a solution WITH the user before any plan is written.
+  2. Load `plan-authoring`. Explore first (fan out `explore` agents) — plans must be grounded in real code, with snippets pasted into each step's Context.
+  3. Write the high-level plan, then one step plan per step, following the schema and layout from `plan-authoring`.
+  4. **Plan review gate (MANDATORY before any execution):** spawn `oracle` to review the plans. Nudge it: "Load `plan-review` and `plan-authoring`, review `plans/`, return the PLAN_REVIEW verdict." REJECT → fix the complaints, re-submit. Do not start execution on an unreviewed or rejected plan.
+  5. Present the reviewed plan to the user for approval.
+
+  ### Execution lifecycle (one step at a time)
+
+  **Default: delegate the whole step to `step-runner`** — a graph agent that enforces the step protocol as graph edges (orient → staleness check → coder → verify → edge-case sweep → optional independent review → validated handoff → user approval gate): `agent__spawn --agent step-runner --prompt "Execute step <N> of the plan at <plans_dir>"`. It returns `STEP_COMPLETE` / `STEP_BLOCKED` / `STEP_REJECTED` / `STEP_FAILED`. Relay its escalations (deviation gate, approval gate) promptly. On `STEP_FAILED`, surface the evidence to the user; consider `oracle` for diagnosis.
+
+  Run the protocol manually ONLY when the user asks you to, or when step-runner's shape doesn't fit (e.g. a docs-only step with nothing to build). Then:
+
+  1. Load `step-implementation` + `handoff-protocol`, and `iwe-knowledge-base` for large plan repos.
+  2. Follow the step protocol phase by phase: orient (previous handoff + `NOTES.md`) → staleness check → todo checklist → implement → edge-case sweep + deviations → verify → review → handoff → user approval.
+  3. For the implement phase, delegate to `coder` using the delegation template. Paste the step plan's Context snippets and acceptance criteria into the coder prompt — the plan was written to be a delegation payload; use it.
+  4. Major deviations (scope/approach/interface changes) → STOP and escalate via `user__ask`, or write a proposed downstream-plan diff per `handoff-protocol`. Never silently absorb them.
+  5. **HARD STOP at the approval gate.** Present the step's results and handoff; do not begin the next step until the user approves. Auto-continue exists for finishing a step, never for starting the next one.
+
+  ## Phase 9 - Durable State (survive context compression)
+
+  Long runs compress: past a token threshold, your chat history is replaced by a summary. Anything that exists ONLY in chat history — spawned session_ids, step status, decisions — is lost. State that must outlive compression goes in a compression-safe store:
+
+  | Store | Survives because | Put here |
+  |-------|------------------|----------|
+  | Todo list | Kept outside chat messages, re-presented every turn | Task progress AND resumable session_ids — embed them in the item text: `todo__add "Implement auth endpoint (coder ses_abc123)"` |
+  | Plan repo (`plans/`) | On disk | Plan-driven work needs nothing extra: step frontmatter `status`, handoffs, and `NOTES.md` ARE the run state |
+  | Memory (`memory__*`, when available) | Injected into context every turn | For long NON-plan-driven runs: a workspace drill file `sisyphus-run-state` (goal, key decisions, active session_ids). Set `expires` to tomorrow; delete it when the run completes |
+
+  Rules:
+
+  1. **Session_ids you may need to resume are never chat-only.** Record them in the todo item for that work the moment the spawn returns. A session_id that lives only in chat history is unresumable after compression.
+  2. **Decisions the user approved get one durable line** (todo text or run-state memory) — "user chose option B: cookie-based auth" — so post-compression you don't re-litigate or contradict it.
+  3. **Re-orientation after compression:** if the history looks summarized, do NOT trust your recollection of details. Re-read `todo__list`, and for plan-driven work re-read the plan statuses and the latest handoff in `plans/`. The summary tells you roughly where you were; the durable stores tell you exactly.
+  4. Do not hoard: run state is not knowledge. Never bloat `MEMORY.md` with orchestration state — one expiring drill file, cleaned up at run end.
+
   ## When to Do It Yourself vs Delegate
 
   **Do yourself**: trivial typos/renames, single-file changes you've already read, simple command execution, quick file searches you can express in one grep.

assets/agents/sisyphus/sbx-mixin.yaml

@@ -0,0 +1,11 @@
+schemaVersion: '1'
+kind: mixin
+name: sisyphus-ddg
+description: >
+  Allows Sisyphus to hit all domains since it utilizes the DuckDuckGo
+  MCP server. This allows the MCP server to actually perform web searches
+  on arbitrary domains and retrieve info for the agent.
+
+network:
+  allowedDomains:
+    - '*'

assets/agents/step-runner/README.md

@@ -0,0 +1,93 @@
+# Step-Runner
+
+A graph-based agent that executes **one step** of a phased implementation
+plan, with the step protocol from the `step-implementation` skill enforced
+as graph edges rather than prose. Designed to be delegated to by
+**[Sisyphus](../sisyphus/README.md)**; delegates implementation to
+**[Coder](../coder/README.md)** and independent review to
+**[code-reviewer](../code-reviewer/README.md)**.
+
+It expects a plan repo authored per the `plan-authoring` skill:
+
+```
+plans/
+  steps/NN-<slug>.md    # step plans with frontmatter (step/title/depends_on/status)
+  handoffs/NN-<slug>.md # written by this agent, validated by a deterministic gate
+  NOTES.md              # rolling durable facts
+```
+
+## Workflow
+
+```
+resolve_step (script)         locate plan + previous handoff, check depends_on,
+        ↓                     mark plan in-progress   [→ gate_blocked if deps unsatisfied]
+orient (llm, read-only)       merge handoff directives + staleness-check the plan
+        ↓
+route_staleness (script)      major deviation → gate_deviation (approval)
+        ↓
+implement (agent → coder)     coder runs its own build/test/self-review fix-loop
+        ↓
+route_coder_result (script)   COMPLETE → verify | REJECTED / FAILED → end
+        ↓
+verify_format_lint (script)   format BEFORE evidence, then lint
+verify_build (script)         step-level build/typecheck
+verify_tests (script)         FULL test suite
+        ↓                     [failures → fix_loop_gate, back-edge to implement]
+edge_case_sweep (llm)         missed edge cases; annotate downstream plans
+        ↓                     (Edge cases sections ONLY - scope changes become proposals)
+route_sweep (script)          5+ files or architectural boundary → independent_review
+independent_review (agent)    code-reviewer; 🔴 findings loop back to implement (bounded)
+        ↓
+write_handoff (llm)           evidence-backed handoff per handoff-protocol + NOTES.md
+check_handoff (script)        deterministic schema gate; marks plan status complete
+        ↓
+gate_user_review (approval)   HARD STOP - approve, or send revision comments
+        ↓                     (revisions loop through implement → verify → handoff again)
+end_success / end_blocked / end_rejected / end_failure
+```
+
+End nodes emit sentinel outcomes for the caller:
+
+- `STEP_COMPLETE` — step implemented, verified, handoff written, user approved.
+- `STEP_BLOCKED` — `depends_on` unsatisfied and the user declined to proceed.
+- `STEP_REJECTED` — user aborted at the deviation gate, or the coder's plan
+  was rejected at its approval gate.
+- `STEP_FAILED` — coder failed, the step-level fix budget was exhausted, or
+  the handoff failed validation twice.
+
+## Usage
+
+```sh
+# From the project root: run the next in-progress/pending step
+coyote -a step-runner "Execute the next step"
+
+# A specific step (also parsed from the prompt: "execute step 3")
+coyote -a step-runner --agent-variable step 3 "Execute step 3"
+
+# Plan repo somewhere else
+coyote -a step-runner --agent-variable plans_dir docs/plans "Execute the next step"
+```
+
+**Invoke from the project root.** The coder sub-agent resolves its own
+`project_dir` from the invocation directory; overriding `project_dir` here
+does not propagate to the spawned coder.
+
+## Tuning
+
+`graph.yaml` `initial_state` exposes:
+
+- `max_fix_attempts` (default `2`) — step-level fix budget (the coder has
+  its own internal budget of 3).
+- `max_review_attempts` (default `1`) — bounded 🔴-finding fix loops after
+  independent review.
+
+Environment overrides honored by the script nodes:
+
+- `FORMAT_CMD` / `LINT_CMD` — formatting and linting (otherwise a per-type
+  heuristic formats, and linting defers to the build/check command).
+- `BUILD_CMD` / `TEST_CMD` — skip project-type detection (same as coder).
+- `STEP_AUTOAPPROVE=1` — bypass the deviation gate (non-interactive runs).
+- `STEP_SKIP_REVIEW=1` — never spawn the independent reviewer.
+
+The final user approval gate is never bypassed by an environment variable -
+it is the point of the workflow.

assets/agents/step-runner/graph.yaml

@@ -0,0 +1,606 @@
+name: step-runner
+description: |
+  Executes ONE step of a phased implementation plan (plans/ repo) with the
+  step protocol enforced as graph edges: orient -> staleness check ->
+  implement (coder) -> verify -> edge-case sweep -> optional independent
+  review -> evidence-backed handoff -> user approval gate. Designed to be
+  delegated to by sisyphus.
+version: '1.0'
+global_tools:
+  - ast_grep.sh
+  - fs_cat.sh
+  - fs_ls.sh
+  - fs_write.sh
+  - fs_patch.sh
+  - execute_command.sh
+
+skills_enabled: true
+enabled_skills:
+  - step-implementation
+  - handoff-protocol
+  - code-review
+  - ai-slop-remover
+
+variables:
+  - name: project_dir
+    description: |
+      Absolute path to the project directory. Defaults to "." (the directory
+      coyote was invoked from). The coder sub-agent resolves its own
+      project_dir the same way, so invoke step-runner FROM the project root
+      unless you override this for both.
+    default: '.'
+  - name: plans_dir
+    description: |
+      Path to the plan repo. Relative paths resolve against project_dir.
+      Expected layout: <plans_dir>/steps/NN-<slug>.md,
+      <plans_dir>/handoffs/, <plans_dir>/NOTES.md.
+    default: 'plans'
+  - name: step
+    description: |
+      Which step to execute: a step number, or "next" to pick the first
+      in-progress (resume) or pending step plan.
+    default: 'next'
+
+settings:
+  max_loop_iterations: 20
+  log_state_snapshots: true
+  validate_before_run: true
+  timeout: 7200
+
+initial_state:
+  project_dir: ''
+  plans_dir: ''
+  step_number: 0
+  step_slug: ''
+  step_title: ''
+  step_plan_path: ''
+  step_plan: ''
+  prev_handoff_path: '(none)'
+  prev_handoff: '(none - this is the first step)'
+  notes_path: ''
+  notes: '(none)'
+  handoff_path: ''
+  blocking_reason: ''
+  plan_summary: ''
+  implementation_brief: ''
+  staleness_report: ''
+  has_major_deviation: false
+  deviation_summary: ''
+  user_feedback: ''
+  fix_instructions: ''
+  fix_attempts: 0
+  max_fix_attempts: 2
+  coder_result: ''
+  format_output: ''
+  lint_ok: true
+  lint_output: ''
+  build_ok: true
+  build_output: ''
+  tests_ok: true
+  tests_output: ''
+  edge_case_report: ''
+  downstream_updates: ''
+  needs_independent_review: false
+  review_report: ''
+  review_attempts: 0
+  max_review_attempts: 1
+  handoff_attempts: 0
+  handoff_fix: ''
+  step_summary: ''
+
+start: resolve_step
+
+nodes:
+  resolve_step:
+    id: resolve_step
+    type: script
+    description: |
+      Locate the step plan, previous handoff, and NOTES.md; parse frontmatter;
+      check depends_on satisfaction against existing handoffs; mark the plan
+      in-progress. Routes to gate_blocked when dependencies are unsatisfied.
+    script: scripts/resolve_step.sh
+    timeout: 30
+    fallback: end_failure
+    next: orient
+
+  gate_blocked:
+    id: gate_blocked
+    type: approval
+    description: Escalate unsatisfied dependencies instead of building on missing ground.
+    question: |
+      Step {{step_number}} ({{step_title}}) is BLOCKED:
+
+      {{blocking_reason}}
+
+      Proceed anyway?
+    options:
+      - 'yes'
+      - 'no'
+    routes:
+      'yes': orient
+      'no': end_blocked
+    on_other: end_blocked
+
+  orient:
+    id: orient
+    type: llm
+    description: |
+      Read-only orientation and staleness check: merge the previous handoff's
+      directives with the step plan, then verify the plan's assumptions
+      against the CURRENT codebase before any edit.
+    skills_enabled: true
+    enabled_skills:
+      - step-implementation
+    instructions: |
+      You are orienting for one step of a phased implementation plan. Load
+      `step-implementation` and apply its Orient and Staleness-check phases.
+      You are READ-ONLY in this node: no edits, no fixes.
+
+      1. Read the previous handoff (below). Note directives aimed at this
+         step, deviations that changed the codebase, and bare assertions
+         that need re-verification.
+      2. Staleness-check the step plan against the code at {{project_dir}}:
+         grep the symbols it references (via execute_command), read its
+         Context snippets at their claimed locations with fs_cat, confirm
+         its Test commands exist.
+      3. Classify discrepancies per the skill's deviation table: minor
+         (mechanics differ; correct silently in the brief) vs major (scope,
+         approach, interfaces, or a later step's assumptions affected).
+
+      Produce `implementation_brief`: the corrected, self-contained marching
+      orders for the implementer - plan tasks in order, handoff directives
+      applied, minor staleness corrections folded in, acceptance criteria
+      restated. The implementer sees ONLY the step plan plus your brief.
+    prompt: |
+      ## Step plan ({{step_plan_path}})
+      {{step_plan}}
+
+      ## Previous handoff ({{prev_handoff_path}})
+      {{prev_handoff}}
+
+      ## Rolling project notes
+      {{notes}}
+    tools:
+      - fs_cat
+      - fs_ls
+      - execute_command
+    max_iterations: 20
+    output_schema:
+      type: object
+      properties:
+        plan_summary:
+          type: string
+          description: 1-3 sentences summarizing what this step delivers
+        implementation_brief:
+          type: string
+          description: Corrected, self-contained instructions for the implementer
+        staleness_report:
+          type: string
+          description: Findings from checking plan assumptions against current code; "clean" if none
+        has_major_deviation:
+          type: boolean
+          description: True when a discrepancy changes scope, approach, or interfaces
+        deviation_summary:
+          type: string
+          description: Major deviations only, with the plan claim vs current reality. Empty when none
+      required:
+        [
+          plan_summary,
+          implementation_brief,
+          staleness_report,
+          has_major_deviation,
+          deviation_summary,
+        ]
+    fallback: end_failure
+    next: route_staleness
+
+  route_staleness:
+    id: route_staleness
+    type: script
+    description: Major deviation -> user gate; otherwise straight to implement.
+    script: scripts/route_staleness.sh
+    timeout: 5
+    fallback: implement
+
+  gate_deviation:
+    id: gate_deviation
+    type: approval
+    description: Major deviations are never silently absorbed - the user decides.
+    question: |
+      Step {{step_number}} ({{step_title}}): the plan no longer matches the
+      codebase in a way that changes scope or approach.
+
+      {{deviation_summary}}
+
+      Staleness report:
+      {{staleness_report}}
+
+      Proceed with the corrected brief? (Answer with anything else to give
+      your own guidance to the implementer.)
+    options:
+      - 'proceed'
+      - 'abort'
+    routes:
+      'proceed': implement
+      'abort': end_rejected
+    on_other: implement
+    state_updates:
+      user_feedback: '{{choice}}'
+
+  implement:
+    id: implement
+    type: agent
+    description: |
+      Delegate implementation to the coder graph agent, which runs its own
+      plan -> implement -> build -> tests -> self-review fix-loop internally.
+    agent: coder
+    prompt: |
+      ## TASK
+      Execute step {{step_number}} ({{step_title}}) of a phased implementation
+      plan for the project at {{project_dir}}.
+
+      ## EXPECTED OUTCOME
+      Every task in the step plan below is implemented and its acceptance
+      criteria are met. Tests are derived from the Acceptance criteria
+      section (not from the implementation). Build and full test suite pass.
+
+      ## MUST DO
+      - Follow the Orientation brief below - it supersedes the raw plan where
+        they disagree (it folds in corrections from the staleness check).
+      - Match the patterns pasted in the step plan's Context section.
+      - Derive tests from the plan's Acceptance criteria.
+
+      ## MUST NOT DO
+      - Do not touch anything listed in the plan's Out of scope section.
+      - Do not modify files under {{plans_dir}}.
+      - Do not implement work belonging to other steps.
+
+      ## CONTEXT
+      ### Step plan
+      {{step_plan}}
+
+      ### Orientation brief (handoff directives + staleness corrections applied)
+      {{implementation_brief}}
+
+      ### User guidance (if any)
+      {{user_feedback}}
+
+      ### Fix loop status (empty on first attempt)
+      {{fix_instructions}}
+    timeout: 3600
+    state_updates:
+      coder_result: '{{output}}'
+    next: route_coder_result
+
+  route_coder_result:
+    id: route_coder_result
+    type: script
+    description: Route on the coder sentinel - COMPLETE verifies, REJECTED/FAILED terminate.
+    script: scripts/route_coder_result.sh
+    timeout: 5
+    fallback: end_failure
+
+  verify_format_lint:
+    id: verify_format_lint
+    type: script
+    description: |
+      Format BEFORE evidence collection (FORMAT_CMD override or per-type
+      heuristic), then lint (LINT_CMD, when configured). Lint failure routes
+      to the fix loop.
+    script: scripts/verify_format_lint.sh
+    timeout: 300
+    fallback: fix_loop_gate
+
+  verify_build:
+    id: verify_build
+    type: script
+    description: Step-level build/typecheck evidence, collected AFTER formatting.
+    script: scripts/verify_build.sh
+    timeout: 600
+    fallback: fix_loop_gate
+
+  verify_tests:
+    id: verify_tests
+    type: script
+    description: FULL test suite - regressions in untouched code fail the step too.
+    script: scripts/verify_tests.sh
+    timeout: 1200
+    fallback: fix_loop_gate
+
+  fix_loop_gate:
+    id: fix_loop_gate
+    type: script
+    description: |
+      Step-level fix budget (the coder already ran its own internal fix
+      loop). Loops to implement with fix_instructions, or ends as failure.
+    script: scripts/fix_loop_gate.sh
+    timeout: 5
+    fallback: end_failure
+
+  edge_case_sweep:
+    id: edge_case_sweep
+    type: llm
+    description: |
+      Post-implementation sweep: missed spots, edge cases, downstream plan
+      implications. May annotate downstream plans' Edge cases sections
+      (annotate vs propose per handoff-protocol). Also judges whether the
+      change warrants an independent review pass.
+    skills_enabled: true
+    enabled_skills:
+      - step-implementation
+      - handoff-protocol
+    instructions: |
+      The implementation for this step just passed build and tests. Load
+      `step-implementation` (edge-case sweep phase) and `handoff-protocol`
+      (annotate-vs-propose rules), then:
+
+      1. Read the changed code (the coder result below names the files).
+         Look for edge cases the plan missed: empty inputs, error paths,
+         concurrency, partial failure, compat.
+      2. For each edge case belonging to a LATER step: check that step's
+         plan under {{plans_dir}}/steps/. If its Edge cases section already
+         covers it, done. If not, append an entry to that section via
+         fs_patch - touch NOTHING else in the file.
+      3. NEVER edit a later plan's Objective, Tasks, Acceptance criteria,
+         or Out of scope. Scope-affecting changes become proposed diffs in
+         `downstream_updates` instead.
+      4. Set needs_independent_review=true when the change touched 5+ files
+         or crosses architectural boundaries (auth, public APIs, schema,
+         security-sensitive paths).
+
+      Be terse. Findings, not prose.
+    prompt: |
+      ## Coder result
+      {{coder_result}}
+
+      ## Step plan
+      {{step_plan}}
+
+      ## Staleness report from orientation
+      {{staleness_report}}
+    tools:
+      - fs_cat
+      - fs_ls
+      - fs_patch
+      - execute_command
+    max_iterations: 20
+    output_schema:
+      type: object
+      properties:
+        edge_case_report:
+          type: string
+          description: Edge cases discovered - both handled and punted, one per line. "none" if empty
+        downstream_updates:
+          type: string
+          description: Annotations made (plan file + section) and proposed diffs for scope-affecting changes. "none" if empty
+        needs_independent_review:
+          type: boolean
+      required: [edge_case_report, downstream_updates, needs_independent_review]
+    fallback: write_handoff
+    next: route_sweep
+
+  route_sweep:
+    id: route_sweep
+    type: script
+    description: Broad or boundary-crossing changes get an independent reviewer.
+    script: scripts/route_sweep.sh
+    timeout: 5
+    fallback: write_handoff
+
+  independent_review:
+    id: independent_review
+    type: agent
+    description: Independent review pass - the author's self-review cannot catch its own rationalizations.
+    agent: code-reviewer
+    prompt: |
+      Review the changes produced for step {{step_number}} ({{step_title}})
+      of a phased implementation plan in {{project_dir}}.
+
+      What the step was supposed to do:
+      {{plan_summary}}
+
+      Coder summary (names the modified/created files):
+      {{coder_result}}
+
+      Review the changed files against the step plan's acceptance criteria.
+      Preserve severity tags in your findings.
+    timeout: 1200
+    state_updates:
+      review_report: '{{output}}'
+    next: route_review
+
+  route_review:
+    id: route_review
+    type: script
+    description: Critical findings loop back to implement (bounded); otherwise proceed to handoff.
+    script: scripts/route_review.sh
+    timeout: 5
+    fallback: write_handoff
+
+  write_handoff:
+    id: write_handoff
+    type: llm
+    description: |
+      Write the evidence-backed handoff per handoff-protocol and append
+      durable facts to NOTES.md. The completion gate (check_handoff)
+      verifies the document afterward.
+    skills_enabled: true
+    enabled_skills:
+      - handoff-protocol
+      - ai-slop-remover
+    instructions: |
+      Load `handoff-protocol` and follow its writer schema EXACTLY: the
+      frontmatter (step, title, result) and all eight sections, writing
+      "None" rather than omitting a section.
+
+      Write the handoff to {{handoff_path}} with fs_write. Paste the
+      verification evidence below verbatim into the Evidence section -
+      commands, exit codes, decisive output lines. Deviations come from the
+      staleness report, gate decisions, and fix loop history. Downstream
+      plan updates come from the sweep results.
+
+      Then append durable, step-independent facts (if any) to {{notes_path}}
+      - create the file if missing, never rewrite existing entries.
+
+      If "Gate feedback" below is non-empty, a previous handoff attempt
+      failed validation - fix exactly what it lists.
+    prompt: |
+      ## Step
+      {{step_number}} ({{step_title}}) - plan at {{step_plan_path}}
+
+      ## Plan summary
+      {{plan_summary}}
+
+      ## Coder result
+      {{coder_result}}
+
+      ## Staleness report / deviations
+      {{staleness_report}}
+
+      Major deviation summary (if any): {{deviation_summary}}
+      User guidance given (if any): {{user_feedback}}
+      Fix loop attempts used: {{fix_attempts}} of {{max_fix_attempts}}
+
+      ## Edge cases discovered
+      {{edge_case_report}}
+
+      ## Downstream plan updates
+      {{downstream_updates}}
+
+      ## Independent review report (if any)
+      {{review_report}}
+
+      ## Verification evidence (paste verbatim)
+      ### Format
+      {{format_output}}
+      ### Lint
+      {{lint_output}}
+      ### Build
+      {{build_output}}
+      ### Tests
+      {{tests_output}}
+
+      ## Gate feedback
+      {{handoff_fix}}
+    tools:
+      - fs_cat
+      - fs_ls
+      - fs_write
+      - fs_patch
+    max_iterations: 15
+    output_schema:
+      type: object
+      properties:
+        step_summary:
+          type: string
+          description: 3-6 sentence summary of the step for the user's approval decision - what was done, deviations, anything needing their attention
+      required: [step_summary]
+    fallback: end_failure
+    next: check_handoff
+
+  check_handoff:
+    id: check_handoff
+    type: script
+    description: |
+      Deterministic completion gate - handoff exists with frontmatter and all
+      required sections. On success, marks the step plan status complete.
+      One retry back to write_handoff, then failure.
+    script: scripts/check_handoff.sh
+    timeout: 10
+    fallback: end_failure
+
+  gate_user_review:
+    id: gate_user_review
+    type: approval
+    description: The hard stop - the next step never starts without explicit approval.
+    question: |
+      ## Step {{step_number}} ({{step_title}}) - ready for review
+
+      {{step_summary}}
+
+      Handoff: {{handoff_path}}
+      Build: {{build_ok}} | Tests: {{tests_ok}} | Fix attempts: {{fix_attempts}}/{{max_fix_attempts}}
+
+      Approve this step? (Answer with anything else to send revision
+      instructions straight to the implementer.)
+    options:
+      - 'approve'
+      - 'revise'
+    routes:
+      'approve': end_success
+      'revise': get_revision
+    on_other: revise_from_choice
+    state_updates:
+      user_feedback: '{{choice}}'
+
+  get_revision:
+    id: get_revision
+    type: input
+    description: Collect revision instructions, then loop back through implement -> verify -> handoff.
+    question: 'What should change? Your comments go to the implementer verbatim.'
+    validation: 'len(input) > 0'
+    state_updates:
+      fix_instructions: '{{input}}'
+    next: implement
+
+  revise_from_choice:
+    id: revise_from_choice
+    type: script
+    description: Free-form approval answers are treated as revision instructions.
+    script: scripts/revise_from_choice.sh
+    timeout: 5
+    fallback: get_revision
+
+  end_success:
+    id: end_success
+    type: end
+    output: |
+      STEP_COMPLETE
+      Step: {{step_number}} ({{step_title}})
+      Plan: {{step_plan_path}}
+      Handoff: {{handoff_path}}
+      Build: passed | Tests: passed | Fix attempts: {{fix_attempts}}/{{max_fix_attempts}}
+
+      {{step_summary}}
+
+      Downstream plan updates:
+      {{downstream_updates}}
+
+  end_blocked:
+    id: end_blocked
+    type: end
+    output: |
+      STEP_BLOCKED
+      Step: {{step_number}} ({{step_title}})
+      Reason:
+      {{blocking_reason}}
+
+  end_rejected:
+    id: end_rejected
+    type: end
+    output: |
+      STEP_REJECTED
+      Step: {{step_number}} ({{step_title}})
+      Rejected at: deviation gate or coder approval gate.
+      Deviation summary:
+      {{deviation_summary}}
+      Coder result (if it ran):
+      {{coder_result}}
+
+  end_failure:
+    id: end_failure
+    type: end
+    output: |
+      STEP_FAILED
+      Step: {{step_number}} ({{step_title}})
+      Fix attempts: {{fix_attempts}}/{{max_fix_attempts}}
+      Blocking reason (if resolution failed): {{blocking_reason}}
+
+      Coder result:
+      {{coder_result}}
+
+      Last build output:
+      {{build_output}}
+
+      Last tests output:
+      {{tests_output}}

assets/agents/step-runner/scripts/check_handoff.sh

@@ -0,0 +1,54 @@
+#!/usr/bin/env bash
+set -uo pipefail
+
+if [[ -n "${GRAPH_STATE_FILE:-}" ]]; then
+  state=$(cat "$GRAPH_STATE_FILE")
+elif [[ -n "${GRAPH_STATE:-}" ]]; then
+  state="$GRAPH_STATE"
+else
+  state='{}'
+fi
+
+handoff_path=$(echo "$state" | jq -r '.handoff_path // ""')
+step_plan_path=$(echo "$state" | jq -r '.step_plan_path // ""')
+handoff_attempts=$(echo "$state" | jq -r '.handoff_attempts // 0')
+
+problems=""
+
+if [[ ! -f "$handoff_path" ]]; then
+  problems="- handoff file does not exist at $handoff_path"$'\n'
+else
+  content=$(cat "$handoff_path")
+  grep -qE '^result:[[:space:]]*(complete|partial|blocked)' <<< "$content" \
+    || problems+="- frontmatter is missing 'result: complete|partial|blocked'"$'\n'
+  for section in "Summary" "Completed" "Not completed" "Deviations" "Downstream plan updates" "Edge cases discovered" "Evidence" "Notes for next step"; do
+    grep -qE "^##[[:space:]]+${section}" <<< "$content" \
+      || problems+="- missing required section: ## ${section}"$'\n'
+  done
+fi
+
+if [[ -z "$problems" ]]; then
+  if [[ -f "$step_plan_path" ]]; then
+    tmp=$(mktemp)
+    awk 'BEGIN{n=0} /^---[[:space:]]*$/{n++; print; next} n==1 && /^status:/{print "status: complete"; next} {print}' "$step_plan_path" > "$tmp" && mv "$tmp" "$step_plan_path"
+  fi
+  jq -nc '{"handoff_fix": "", "_next": "gate_user_review"}'
+  exit 0
+fi
+
+if (( handoff_attempts >= 1 )); then
+  jq -nc \
+    --arg br "Handoff failed validation twice. Problems:
+$problems" \
+    '{"blocking_reason": $br, "_next": "end_failure"}'
+  exit 0
+fi
+
+jq -nc \
+  --arg hf "The previous handoff attempt failed validation. Fix exactly these problems:
+$problems" \
+  '{
+    "handoff_attempts": 1,
+    "handoff_fix": $hf,
+    "_next": "write_handoff"
+  }'

assets/agents/step-runner/scripts/fix_loop_gate.sh

@@ -0,0 +1,60 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+if [[ -n "${GRAPH_STATE_FILE:-}" ]]; then
+  state=$(cat "$GRAPH_STATE_FILE")
+elif [[ -n "${GRAPH_STATE:-}" ]]; then
+  state="$GRAPH_STATE"
+else
+  state='{}'
+fi
+
+fix_attempts=$(echo "$state" | jq -r '.fix_attempts // 0')
+max_fix_attempts=$(echo "$state" | jq -r '.max_fix_attempts // 2')
+lint_ok=$(echo "$state" | jq -r '.lint_ok | if . == null then "true" else (. | tostring) end')
+build_ok=$(echo "$state" | jq -r '.build_ok | if . == null then "true" else (. | tostring) end')
+tests_ok=$(echo "$state" | jq -r '.tests_ok | if . == null then "true" else (. | tostring) end')
+lint_output=$(echo "$state" | jq -r '.lint_output // ""')
+build_output=$(echo "$state" | jq -r '.build_output // ""')
+tests_output=$(echo "$state" | jq -r '.tests_output // ""')
+
+if (( fix_attempts >= max_fix_attempts )); then
+  jq -nc \
+    --argjson n "$fix_attempts" \
+    '{
+      "fix_attempts": $n,
+      "_next": "end_failure"
+    }'
+  exit 0
+fi
+
+next_attempts=$((fix_attempts + 1))
+
+if [[ "$lint_ok" != "true" ]]; then
+  stage="lint"
+  output="$lint_output"
+elif [[ "$build_ok" != "true" ]]; then
+  stage="build"
+  output="$build_output"
+elif [[ "$tests_ok" != "true" ]]; then
+  stage="full test suite"
+  output="$tests_output"
+else
+  stage="verification"
+  output="fix_loop_gate was reached but no failing stage was recorded. Re-run verification."
+fi
+
+fix_instructions=$(printf '## Fix loop status (step-level attempt %d of %d)\n\nThe implementation passed the coder'"'"'s internal checks but failed step-level verification at the %s stage.\n\nOutput:\n```\n%s\n```\n\nIdentify the minimal fix and apply it. Do not refactor. Regressions in untouched code caused by this change are in scope.' \
+  "$next_attempts" "$max_fix_attempts" "$stage" "$output")
+
+jq -nc \
+  --argjson n "$next_attempts" \
+  --arg 'fi' "$fix_instructions" \
+  '{
+    "fix_attempts": $n,
+    "fix_instructions": $fi,
+    "lint_ok": true,
+    "build_ok": true,
+    "tests_ok": true,
+    "_next": "implement"
+  }'

assets/agents/step-runner/scripts/resolve_step.sh

@@ -0,0 +1,152 @@
+#!/usr/bin/env bash
+set -uo pipefail
+
+if [[ -n "${GRAPH_STATE_FILE:-}" ]]; then
+  state=$(cat "$GRAPH_STATE_FILE")
+elif [[ -n "${GRAPH_STATE:-}" ]]; then
+  state="$GRAPH_STATE"
+else
+  state='{}'
+fi
+
+fail() {
+  jq -nc --arg r "$1" '{"blocking_reason": $r, "_next": "end_failure"}'
+  exit 0
+}
+
+project_dir="${LLM_AGENT_VAR_PROJECT_DIR:-.}"
+project_dir=$(cd "$project_dir" 2>/dev/null && pwd) || fail "project_dir does not exist: $project_dir"
+
+plans_dir="${LLM_AGENT_VAR_PLANS_DIR:-plans}"
+[[ "$plans_dir" != /* ]] && plans_dir="$project_dir/$plans_dir"
+steps_dir="$plans_dir/steps"
+handoffs_dir="$plans_dir/handoffs"
+notes_path="$plans_dir/NOTES.md"
+
+[[ -d "$steps_dir" ]] || fail "No step plans directory at $steps_dir (expected <plans_dir>/steps/NN-<slug>.md)"
+
+frontmatter() {
+  awk '/^---[[:space:]]*$/{n++; next} n==1{print} n>=2{exit}' "$1"
+}
+
+fm_value() {
+  echo "$1" | grep -E "^$2:" | head -1 | sed -E "s/^$2:[[:space:]]*//" | sed -E 's/^["'"'"']|["'"'"']$//g'
+}
+
+step="${LLM_AGENT_VAR_STEP:-next}"
+if [[ "$step" == "next" ]]; then
+  prompt_step=$(echo "$state" | jq -r '.initial_prompt // ""' | grep -oiE 'step[[:space:]#:]*[0-9]+' | head -1 | grep -oE '[0-9]+' || true)
+  [[ -n "$prompt_step" ]] && step="$prompt_step"
+fi
+
+plan_file=""
+if [[ "$step" == "next" ]]; then
+  first_pending=""
+  while IFS= read -r f; do
+    st=$(fm_value "$(frontmatter "$f")" "status")
+    if [[ "$st" == "in-progress" ]]; then
+      plan_file="$f"
+      break
+    fi
+    [[ -z "$first_pending" && ( "$st" == "pending" || -z "$st" ) ]] && first_pending="$f"
+  done < <(find "$steps_dir" -maxdepth 1 -name '*.md' | sort)
+  [[ -z "$plan_file" ]] && plan_file="$first_pending"
+  [[ -z "$plan_file" ]] && fail "No in-progress or pending step plans in $steps_dir"
+else
+  [[ "$step" =~ ^[0-9]+$ ]] || fail "step must be a number or 'next'; got: $step"
+  padded=$(printf '%02d' "$((10#$step))")
+  plan_file=$(find "$steps_dir" -maxdepth 1 \( -name "${padded}-*.md" -o -name "${step}-*.md" \) | sort | head -1)
+  [[ -n "$plan_file" ]] || fail "No step plan matching step $step in $steps_dir"
+fi
+
+bn=$(basename "$plan_file" .md)
+num_part="${bn%%-*}"
+[[ "$num_part" =~ ^[0-9]+$ ]] || fail "Step plan filename must start with a number: $bn"
+step_number=$((10#$num_part))
+step_slug="${bn#*-}"
+
+fm=$(frontmatter "$plan_file")
+step_title=$(fm_value "$fm" "title")
+[[ -z "$step_title" ]] && step_title="$step_slug"
+
+deps=$(echo "$fm" | awk '/^depends_on:/{f=1; print; next} f && /^[[:space:]]*-/{print; next} f{exit}' | grep -oE '[0-9]+' || true)
+unsatisfied=""
+for dep in $deps; do
+  dep_padded=$(printf '%02d' "$((10#$dep))")
+  dep_handoff=$(find "$handoffs_dir" -maxdepth 1 \( -name "${dep_padded}-*.md" -o -name "${dep}-*.md" \) 2>/dev/null | sort | head -1)
+  if [[ -z "$dep_handoff" ]]; then
+    unsatisfied+="- step $dep: no handoff found (step not executed?)"$'\n'
+    continue
+  fi
+  dep_result=$(fm_value "$(frontmatter "$dep_handoff")" "result")
+  if [[ "$dep_result" != "complete" ]]; then
+    unsatisfied+="- step $dep: handoff result is '$dep_result' (not complete): $dep_handoff"$'\n'
+  fi
+done
+
+prev_handoff_path="(none)"
+prev_handoff="(none - this is the first step)"
+prev_file=""
+prev_num=0
+while IFS= read -r h; do
+  hn="${h##*/}"
+  hn="${hn%%-*}"
+  [[ "$hn" =~ ^[0-9]+$ ]] || continue
+  n=$((10#$hn))
+  if (( n < step_number && n >= prev_num )); then
+    prev_num=$n
+    prev_file="$h"
+  fi
+done < <(find "$handoffs_dir" -maxdepth 1 -name '*.md' 2>/dev/null | sort)
+if [[ -n "$prev_file" ]]; then
+  prev_handoff_path="$prev_file"
+  prev_handoff=$(head -c 16000 "$prev_file")
+fi
+
+notes="(none)"
+[[ -f "$notes_path" ]] && notes=$(head -c 8000 "$notes_path")
+
+step_plan=$(head -c 24000 "$plan_file")
+handoff_path="$handoffs_dir/$(basename "$plan_file")"
+
+tmp=$(mktemp)
+awk 'BEGIN{n=0} /^---[[:space:]]*$/{n++; print; next} n==1 && /^status:/{print "status: in-progress"; next} {print}' "$plan_file" > "$tmp" && mv "$tmp" "$plan_file"
+
+next_node="orient"
+blocking_reason=""
+if [[ -n "$unsatisfied" ]]; then
+  next_node="gate_blocked"
+  blocking_reason="Unsatisfied dependencies:"$'\n'"$unsatisfied"
+fi
+
+jq -nc \
+  --arg pd "$project_dir" \
+  --arg pl "$plans_dir" \
+  --argjson sn "$step_number" \
+  --arg ss "$step_slug" \
+  --arg st "$step_title" \
+  --arg spp "$plan_file" \
+  --arg sp "$step_plan" \
+  --arg php "$prev_handoff_path" \
+  --arg ph "$prev_handoff" \
+  --arg np "$notes_path" \
+  --arg no "$notes" \
+  --arg hp "$handoff_path" \
+  --arg br "$blocking_reason" \
+  --arg nx "$next_node" \
+  '{
+    "project_dir": $pd,
+    "plans_dir": $pl,
+    "step_number": $sn,
+    "step_slug": $ss,
+    "step_title": $st,
+    "step_plan_path": $spp,
+    "step_plan": $sp,
+    "prev_handoff_path": $php,
+    "prev_handoff": $ph,
+    "notes_path": $np,
+    "notes": $no,
+    "handoff_path": $hp,
+    "blocking_reason": $br,
+    "_next": $nx
+  }'

assets/agents/step-runner/scripts/revise_from_choice.sh

@@ -0,0 +1,27 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+if [[ -n "${GRAPH_STATE_FILE:-}" ]]; then
+  state=$(cat "$GRAPH_STATE_FILE")
+elif [[ -n "${GRAPH_STATE:-}" ]]; then
+  state="$GRAPH_STATE"
+else
+  state='{}'
+fi
+
+feedback=$(echo "$state" | jq -r '.user_feedback // ""')
+
+if [[ -z "$feedback" ]]; then
+  jq -nc '{"_next": "get_revision"}'
+  exit 0
+fi
+
+fix_instructions=$(printf '## Revision requested by the user at the step approval gate\n\nAddress these comments with minimal edits, then the step re-verifies and the handoff is rewritten:\n\n%s' \
+  "$feedback")
+
+jq -nc \
+  --arg 'fi' "$fix_instructions" \
+  '{
+    "fix_instructions": $fi,
+    "_next": "implement"
+  }'

assets/agents/step-runner/scripts/route_coder_result.sh

@@ -0,0 +1,27 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+if [[ -n "${GRAPH_STATE_FILE:-}" ]]; then
+  state=$(cat "$GRAPH_STATE_FILE")
+elif [[ -n "${GRAPH_STATE:-}" ]]; then
+  state="$GRAPH_STATE"
+else
+  state='{}'
+fi
+
+coder_result=$(echo "$state" | jq -r '.coder_result // ""')
+
+case "$coder_result" in
+  *CODER_COMPLETE*)
+    jq -nc '{"_next": "verify_format_lint"}'
+    ;;
+  *CODER_REJECTED*)
+    jq -nc '{"_next": "end_rejected"}'
+    ;;
+  *CODER_FAILED*)
+    jq -nc '{"blocking_reason": "coder fix-loop exhausted; see coder result", "_next": "end_failure"}'
+    ;;
+  *)
+    jq -nc '{"blocking_reason": "coder returned no recognizable sentinel (expected CODER_COMPLETE / CODER_REJECTED / CODER_FAILED)", "_next": "end_failure"}'
+    ;;
+esac

assets/agents/step-runner/scripts/route_review.sh

@@ -0,0 +1,38 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+if [[ -n "${GRAPH_STATE_FILE:-}" ]]; then
+  state=$(cat "$GRAPH_STATE_FILE")
+elif [[ -n "${GRAPH_STATE:-}" ]]; then
+  state="$GRAPH_STATE"
+else
+  state='{}'
+fi
+
+review_report=$(echo "$state" | jq -r '.review_report // ""')
+review_attempts=$(echo "$state" | jq -r '.review_attempts // 0')
+max_review_attempts=$(echo "$state" | jq -r '.max_review_attempts // 1')
+
+if ! grep -qF "🔴" <<< "$review_report"; then
+  jq -nc '{"_next": "write_handoff"}'
+  exit 0
+fi
+
+if (( review_attempts >= max_review_attempts )); then
+  jq -nc '{"_next": "write_handoff"}'
+  exit 0
+fi
+
+next_review=$((review_attempts + 1))
+fix_instructions=$(printf '## Independent review findings (attempt %d of %d)\n\nAn independent reviewer flagged CRITICAL (🔴) findings. Address ONLY the 🔴 findings with minimal edits. Do not refactor unrelated code.\n\n%s' \
+  "$next_review" "$max_review_attempts" "$review_report")
+
+jq -nc \
+  --argjson n "$next_review" \
+  --arg 'fi' "$fix_instructions" \
+  '{
+    "review_attempts": $n,
+    "fix_instructions": $fi,
+    "needs_independent_review": false,
+    "_next": "implement"
+  }'

assets/agents/step-runner/scripts/route_staleness.sh

@@ -0,0 +1,23 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+if [[ -n "${GRAPH_STATE_FILE:-}" ]]; then
+  state=$(cat "$GRAPH_STATE_FILE")
+elif [[ -n "${GRAPH_STATE:-}" ]]; then
+  state="$GRAPH_STATE"
+else
+  state='{}'
+fi
+
+has_major=$(echo "$state" | jq -r '.has_major_deviation // false')
+
+if [[ "${STEP_AUTOAPPROVE:-0}" == "1" ]]; then
+  jq -nc '{"_next": "implement"}'
+  exit 0
+fi
+
+if [[ "$has_major" == "true" ]]; then
+  jq -nc '{"_next": "gate_deviation"}'
+else
+  jq -nc '{"_next": "implement"}'
+fi

assets/agents/step-runner/scripts/route_sweep.sh

@@ -0,0 +1,23 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+if [[ -n "${GRAPH_STATE_FILE:-}" ]]; then
+  state=$(cat "$GRAPH_STATE_FILE")
+elif [[ -n "${GRAPH_STATE:-}" ]]; then
+  state="$GRAPH_STATE"
+else
+  state='{}'
+fi
+
+needs_review=$(echo "$state" | jq -r '.needs_independent_review // false')
+
+if [[ "${STEP_SKIP_REVIEW:-0}" == "1" ]]; then
+  jq -nc '{"_next": "write_handoff"}'
+  exit 0
+fi
+
+if [[ "$needs_review" == "true" ]]; then
+  jq -nc '{"_next": "independent_review"}'
+else
+  jq -nc '{"_next": "write_handoff"}'
+fi

assets/agents/step-runner/scripts/verify_build.sh

@@ -0,0 +1,57 @@
+#!/usr/bin/env bash
+set -uo pipefail
+
+# shellcheck disable=SC1091
+source "$(dirname "$0")/../../.shared/utils.sh"
+
+if [[ -n "${GRAPH_STATE_FILE:-}" ]]; then
+  state=$(cat "$GRAPH_STATE_FILE")
+elif [[ -n "${GRAPH_STATE:-}" ]]; then
+  state="$GRAPH_STATE"
+else
+  state='{}'
+fi
+
+project_dir=$(echo "$state" | jq -r '.project_dir // "."')
+
+if [[ -n "${BUILD_CMD:-}" ]]; then
+  cmd="$BUILD_CMD"
+else
+  project_info=$(detect_project "$project_dir")
+  cmd=$(echo "$project_info" | jq -r '.check // .build // ""')
+fi
+
+if [[ -z "$cmd" || "$cmd" == "null" ]]; then
+  jq -nc '{
+    "build_ok": true,
+    "build_output": "(no build/check command available for this project type)",
+    "_next": "verify_tests"
+  }'
+  exit 0
+fi
+
+exit_code=0
+output=$(cd "$project_dir" && eval "$cmd" 2>&1) || exit_code=$?
+
+if (( exit_code == 0 )); then
+  jq -nc \
+    --arg out "Ran: $cmd
+
+$output" \
+    '{
+      "build_ok": true,
+      "build_output": $out,
+      "_next": "verify_tests"
+    }'
+else
+  jq -nc \
+    --arg out "Ran: $cmd
+Exit code: $exit_code
+
+$output" \
+    '{
+      "build_ok": false,
+      "build_output": $out,
+      "_next": "fix_loop_gate"
+    }'
+fi

assets/agents/step-runner/scripts/verify_format_lint.sh

@@ -0,0 +1,79 @@
+#!/usr/bin/env bash
+set -uo pipefail
+
+# shellcheck disable=SC1091
+source "$(dirname "$0")/../../.shared/utils.sh"
+
+if [[ -n "${GRAPH_STATE_FILE:-}" ]]; then
+  state=$(cat "$GRAPH_STATE_FILE")
+elif [[ -n "${GRAPH_STATE:-}" ]]; then
+  state="$GRAPH_STATE"
+else
+  state='{}'
+fi
+
+project_dir=$(echo "$state" | jq -r '.project_dir // "."')
+project_type=$(detect_project "$project_dir" | jq -r '.type // "unknown"')
+
+format_cmd="${FORMAT_CMD:-}"
+if [[ -z "$format_cmd" ]]; then
+  case "$project_type" in
+    rust) format_cmd="cargo fmt" ;;
+    go) format_cmd="gofmt -w ." ;;
+    python) command -v ruff &>/dev/null && format_cmd="ruff format ." ;;
+  esac
+fi
+
+if [[ -z "$format_cmd" ]]; then
+  format_output="(no format command configured for project type '$project_type'; skipped. Set FORMAT_CMD to enable.)"
+else
+  fmt_rc=0
+  fmt_out=$(cd "$project_dir" && eval "$format_cmd" 2>&1) || fmt_rc=$?
+  format_output="Ran: $format_cmd
+Exit code: $fmt_rc
+
+$fmt_out"
+fi
+
+lint_cmd="${LINT_CMD:-}"
+if [[ -z "$lint_cmd" ]]; then
+  jq -nc \
+    --arg fo "$format_output" \
+    '{
+      "format_output": $fo,
+      "lint_ok": true,
+      "lint_output": "(no LINT_CMD configured; linting is covered by the build/check command)",
+      "_next": "verify_build"
+    }'
+  exit 0
+fi
+
+lint_rc=0
+lint_out=$(cd "$project_dir" && eval "$lint_cmd" 2>&1) || lint_rc=$?
+
+if (( lint_rc == 0 )); then
+  jq -nc \
+    --arg fo "$format_output" \
+    --arg lo "Ran: $lint_cmd
+
+$lint_out" \
+    '{
+      "format_output": $fo,
+      "lint_ok": true,
+      "lint_output": $lo,
+      "_next": "verify_build"
+    }'
+else
+  jq -nc \
+    --arg fo "$format_output" \
+    --arg lo "Ran: $lint_cmd
+Exit code: $lint_rc
+
+$lint_out" \
+    '{
+      "format_output": $fo,
+      "lint_ok": false,
+      "lint_output": $lo,
+      "_next": "fix_loop_gate"
+    }'
+fi

assets/agents/step-runner/scripts/verify_tests.sh

@@ -0,0 +1,57 @@
+#!/usr/bin/env bash
+set -uo pipefail
+
+# shellcheck disable=SC1091
+source "$(dirname "$0")/../../.shared/utils.sh"
+
+if [[ -n "${GRAPH_STATE_FILE:-}" ]]; then
+  state=$(cat "$GRAPH_STATE_FILE")
+elif [[ -n "${GRAPH_STATE:-}" ]]; then
+  state="$GRAPH_STATE"
+else
+  state='{}'
+fi
+
+project_dir=$(echo "$state" | jq -r '.project_dir // "."')
+
+if [[ -n "${TEST_CMD:-}" ]]; then
+  cmd="$TEST_CMD"
+else
+  project_info=$(detect_project "$project_dir")
+  cmd=$(echo "$project_info" | jq -r '.test // ""')
+fi
+
+if [[ -z "$cmd" || "$cmd" == "null" ]]; then
+  jq -nc '{
+    "tests_ok": true,
+    "tests_output": "(no test command available for this project type)",
+    "_next": "edge_case_sweep"
+  }'
+  exit 0
+fi
+
+exit_code=0
+output=$(cd "$project_dir" && eval "$cmd" 2>&1) || exit_code=$?
+
+if (( exit_code == 0 )); then
+  jq -nc \
+    --arg out "Ran: $cmd
+
+$output" \
+    '{
+      "tests_ok": true,
+      "tests_output": $out,
+      "_next": "edge_case_sweep"
+    }'
+else
+  jq -nc \
+    --arg out "Ran: $cmd
+Exit code: $exit_code
+
+$output" \
+    '{
+      "tests_ok": false,
+      "tests_output": $out,
+      "_next": "fix_loop_gate"
+    }'
+fi

assets/functions/mcp.json

@@ -18,6 +18,11 @@
       "type": "stdio",
       "command": "uvx",
       "args": ["duckduckgo-mcp-server"]
+    },
+    "iwe": {
+      "type": "stdio",
+      "command": "iwec",
+      "args": ["--project", "."]
     }
   }
 }

assets/functions/sbx-mixin.yaml

@@ -0,0 +1,44 @@
+schemaVersion: "1"
+kind: mixin
+name: built-in-tools
+description: >
+  Installs binaries and allows network domains required by Coyote's built-in
+  global tools and the default MCP server set. Auto-applied by Coyote's sbx
+  mixin discovery when running `coyote --sandbox`.
+
+network:
+  allowedDomains:
+    # fetch_url_via_jina + jina reader fallback
+    - "r.jina.ai:443"
+    # get_current_weather (.sh, .py, .ts)
+    - "wttr.in:443"
+    # search_arxiv (the .sh tool still uses http://, so :80 is required until fixed)
+    - "export.arxiv.org:443"
+    - "export.arxiv.org:80"
+    # search_arxiv + search_wikipedia may follow DOI redirects
+    - "doi.org:443"
+    # search_wikipedia
+    - "en.wikipedia.org:443"
+    # search_wolframalpha
+    - "api.wolframalpha.com:443"
+    # web_search_perplexity
+    - "api.perplexity.ai:443"
+    # web_search_tavily
+    - "api.tavily.com:443"
+    # send_twilio
+    - "api.twilio.com:443"
+    # MCP: github (built-in mcp.json: api.githubcopilot.com)
+    - "api.githubcopilot.com:443"
+    # MCP: atlassian (built-in mcp.json: mcp-remote -> mcp.atlassian.com)
+    - "mcp.atlassian.com:443"
+    # MCP: ddg-search (built-in mcp.json: uvx duckduckgo-mcp-server)
+    - "duckduckgo.com:443"
+    - "html.duckduckgo.com:443"
+    - "lite.duckduckgo.com:443"
+    # MCP: npx-based servers (mcp-remote) pull from npm
+    - "registry.npmjs.org:443"
+    # MCP: docker server may pull images from common registries
+    - "ghcr.io:443"
+    - "registry-1.docker.io:443"
+    - "auth.docker.io:443"
+    - "production.cloudflare.docker.com:443"

assets/functions/tools/ast_grep.sh

@@ -0,0 +1,81 @@
+#!/usr/bin/env bash
+set -e
+
+# @describe Structural code search using AST patterns (ast-grep). Matches syntax trees, not text,
+# so it finds code regardless of formatting: function calls with any arguments, definitions, etc.
+# Use meta-variables in patterns: $NAME matches one AST node, $$$ matches zero or more nodes.
+# Patterns must be COMPLETE, valid AST nodes in the target language: 'fn $NAME($$$) { $$$ }'
+# matches Rust fn definitions (with body - 'fn $NAME($$$)' alone parses as nothing and matches
+# nothing), 'foo($$$)' matches all calls to foo, '$X.unwrap()' matches all unwrap calls.
+# Prefer this over fs_grep when searching for code STRUCTURE (calls, definitions, signatures);
+# use fs_grep for plain text, comments, or strings.
+
+# @option --pattern! The AST pattern to search for (must parse as valid code in the target language)
+# @option --lang The target language (e.g. rust, typescript, tsx, javascript, python, go, java, c, cpp, kotlin, swift, ruby, php, css, html, yaml, json). Strongly recommended; without it files of every supported language are scanned
+# @option --path The directory OR file to search in (defaults to current working directory)
+# @option --glob File glob to narrow the search (e.g. "src/**/*.rs", "!**/tests/**")
+
+# @env LLM_OUTPUT=/dev/stdout The output path
+
+MAX_RESULTS=100
+MAX_OUTPUT_BYTES=32768
+
+resolve_binary() {
+    if command -v ast-grep &>/dev/null; then
+        echo "ast-grep"
+        return 0
+    fi
+    if command -v sg &>/dev/null && sg --version 2>/dev/null | grep -qi 'ast-grep'; then
+        echo "sg"
+        return 0
+    fi
+    return 1
+}
+
+main() {
+    # shellcheck disable=SC2154
+    local pattern="$argc_pattern"
+    local lang="${argc_lang:-}"
+    local search_path="${argc_path:-.}"
+    local glob="${argc_glob:-}"
+
+    local bin
+    if ! bin=$(resolve_binary); then
+        printf 'ast-grep is not installed. Fall back to fs_grep for this search.\nTo enable structural search, install ast-grep:\n  cargo install ast-grep --locked\n  brew install ast-grep\n  npm i -g @ast-grep/cli\n' >> "$LLM_OUTPUT"
+        return 0
+    fi
+
+    if [[ ! -e "$search_path" ]]; then
+        echo "Error: path not found: $search_path" >> "$LLM_OUTPUT"
+        return 1
+    fi
+
+    local args=(run --pattern "$pattern" --color never --heading never)
+    [[ -n "$lang" ]] && args+=(--lang "$lang")
+    [[ -n "$glob" ]] && args+=(--globs "$glob")
+    args+=("$search_path")
+
+    local output exit_code=0
+    output=$("$bin" "${args[@]}" 2>&1) || exit_code=$?
+
+    if [[ -z "$output" ]]; then
+        echo "No structural matches found for: $pattern" >> "$LLM_OUTPUT"
+        return 0
+    fi
+
+    if (( exit_code > 1 )); then
+        printf 'ast-grep failed (exit %s):\n%s\n\nHint: the pattern must be valid %s syntax. Meta-variables: $NAME (one node), $$$ (zero or more).\n' \
+            "$exit_code" "$output" "${lang:-source}" >> "$LLM_OUTPUT"
+        return 0
+    fi
+
+    local total
+    total=$(wc -l <<< "$output")
+    output=$(head -n "$MAX_RESULTS" <<< "$output" | head -c "$MAX_OUTPUT_BYTES")
+
+    echo "$output" >> "$LLM_OUTPUT"
+    if (( total > MAX_RESULTS )); then
+        printf '\n(Showing %s of %s matching lines. Narrow with --glob, --lang, or a more specific pattern.)\n' \
+            "$MAX_RESULTS" "$total" >> "$LLM_OUTPUT"
+    fi
+}

assets/functions/tools/fs_cat.sh

@@ -10,5 +10,13 @@ set -e
 
 main() {
     # shellcheck disable=SC2154
-		cat "$argc_path" >> "$LLM_OUTPUT" 2>&1 || echo "No such file or path: $argc_path" >> "$LLM_OUTPUT"
+    local path="$argc_path"
+
+    # An empty result is shown to the model as the opaque literal "DONE"; emit a note instead.
+    if [[ -f "$path" && ! -s "$path" ]]; then
+        echo "(empty file: $path)" >> "$LLM_OUTPUT"
+        return 0
+    fi
+
+    cat "$path" >> "$LLM_OUTPUT" 2>&1 || echo "No such file or path: $path" >> "$LLM_OUTPUT"
 }

assets/functions/tools/fs_glob.sh

@@ -17,8 +17,8 @@ main() {
     local search_path="${argc_path:-.}"
 
     if [[ ! -d "$search_path" ]]; then
-        echo "Error: directory not found: $search_path" >> "$LLM_OUTPUT"
+        echo "Error: directory not found: $search_path" >&2
-        return 1
+        exit 1
     fi
 
     local results

assets/functions/tools/fs_grep.sh

@@ -21,8 +21,8 @@ main() {
     local include_filter="${argc_include:-}"
 
     if [[ ! -e "$search_path" ]]; then
-        echo "Error: path not found: $search_path" >> "$LLM_OUTPUT"
+        echo "Error: path not found: $search_path" >&2
-        return 1
+        exit 1
     fi
 
     local grep_args=(-nH --color=never)

assets/functions/tools/fs_ls.sh

@@ -9,5 +9,18 @@ set -e
 
 main() {
     # shellcheck disable=SC2154
-    ls -1 "$argc_path" >> "$LLM_OUTPUT" 2>&1 || echo "No such path: $argc_path" >> "$LLM_OUTPUT"
+    local path="$argc_path"
+    local output
+
+    if ! output=$(ls -1 "$path" 2>&1); then
+        echo "$output" >> "$LLM_OUTPUT"
+        return 0
+    fi
+
+    # An empty result is shown to the model as the opaque literal "DONE"; emit a note instead.
+    if [[ -z "$output" ]]; then
+        echo "(empty directory: $path)" >> "$LLM_OUTPUT"
+    else
+        echo "$output" >> "$LLM_OUTPUT"
+    fi
 }

assets/functions/tools/fs_patch.sh

@@ -5,6 +5,23 @@ set -e
 # PREFERRED way to modify a file. Prefer this over fs_write whenever the file already exists: it sends less data,
 # preserves unchanged content automatically, and is less prone to accidental data loss from full rewrites.
 # Use fs_write only when you are creating a new file or doing a complete rewrite where most of the content changes.
+#
+# CRITICAL — the patch is matched byte-for-byte. There is no fuzzy matching, no whitespace tolerance, and no context shift:
+# - Context lines (prefixed with a single space) and removed lines (prefixed with '-') must equal the file content exactly.
+#   If unsure, fs_cat the file first and copy the bytes verbatim into your patch.
+# - JSON-escape the contents string ONCE. Each literal backslash in the file becomes \\ in the JSON contents string. So a
+#   shell line containing s|\\"|"|g must appear in JSON as s|\\\\\"|\"|g — NOT s|\\\\\\\"|\\\"|g. Over-escaping backslashes
+#   is the most common cause of "unable to apply patch" failures, especially in files with sed/jq/regex pipelines or
+#   embedded Python with quoted strings.
+# - Hunks are applied in order; the first hunk that fails aborts the whole patch — later hunks are NOT attempted.
+# - If you've edited this file in earlier tool calls, fs_cat it again before composing the patch. A stale view of the file
+#   produces context lines that no longer match.
+# - On failure the error message names the failing hunk and shows the expected-vs-actual line. Fix that specific line and
+#   retry — do not blindly resend a near-identical patch.
+#
+# For files with heavy escaping (sed/jq/regex pipelines, shell with embedded heredocs, deeply quoted strings), prefer
+# fs_write over chained fs_patch hunks to replace the entire file with the full new contents (i.e. original content +
+# your changes).
 
 # @option --path! The path of the file to apply the patch to
 # @option --contents! The patch to apply to the file

assets/functions/tools/fs_read.sh

@@ -8,8 +8,8 @@ set -e
 # Use the grep tool to find specific content before reading, then read with offset to target the relevant section.
 
 # @option --path! The absolute path to the file or directory to read
-# @option --offset The line number to start reading from (1-indexed, default: 1)
+# @option --offset <INT> The line number to start reading from (1-indexed, default: 1)
-# @option --limit The maximum number of lines to read (default: 2000)
+# @option --limit <INT> The maximum number of lines to read (default: 2000)
 
 # @env LLM_OUTPUT=/dev/stdout The output path
 
@@ -23,8 +23,8 @@ main() {
     local limit="${argc_limit:-2000}"
 
     if [[ ! -e "$target" ]]; then
-        echo "Error: path not found: $target" >> "$LLM_OUTPUT"
+        echo "Error: path not found: $target" >&2
-        return 1
+        exit 1
     fi
 
     if [[ -d "$target" ]]; then
@@ -33,9 +33,20 @@ main() {
     fi
 
     local total_lines file_bytes
-    total_lines=$(wc -l < "$target" 2>/dev/null || echo 0)
+    # awk counts a final line that lacks a trailing newline; wc -l would undercount it by one.
+    total_lines=$(awk 'END { print NR }' "$target" 2>/dev/null || echo 0)
     file_bytes=$(wc -c < "$target" 2>/dev/null || echo 0)
 
+    if [[ "$total_lines" -eq 0 ]]; then
+        echo "(file is empty: $target)" >> "$LLM_OUTPUT"
+        return 0
+    fi
+
+    if [[ "$offset" -gt "$total_lines" ]]; then
+        echo "(offset $offset is past the end of the file, which has $total_lines lines)" >> "$LLM_OUTPUT"
+        return 0
+    fi
+
     if [[ "$file_bytes" -gt "$MAX_BYTES" ]] && [[ "$offset" -eq 1 ]] && [[ "$limit" -ge 2000 ]]; then
         {
             echo "Warning: Large file (${file_bytes} bytes, ${total_lines} lines). Showing first ${limit} lines."
@@ -48,7 +59,8 @@ main() {
 
     sed -n "${offset},${end_line}p" "$target" 2>/dev/null | {
         local line_num=$offset
-        while IFS= read -r line; do
+        # `|| [[ -n "$line" ]]` keeps the final line when the file has no trailing newline.
+        while IFS= read -r line || [[ -n "$line" ]]; do
             if [[ ${#line} -gt $MAX_LINE_LENGTH ]]; then
                 line="${line:0:$MAX_LINE_LENGTH}... (truncated)"
             fi

assets/functions/utils/prompt-utils.sh

@@ -507,7 +507,9 @@ open_link() {
 
 guard_operation() {
 	if [[ -z "$AUTO_CONFIRM" && -z "$LLM_AGENT_VAR_AUTO_CONFIRM" ]]; then
-		ans="$(confirm "${1:-Are you sure you want to continue?}")"
+		# 2>/dev/tty: keep the prompt off the host-captured stderr pipe so it
+		# can't leak into tool_call_error JSON when the wrapped command fails.
+		ans="$(confirm "${1:-Are you sure you want to continue?}" 2>/dev/tty)"
 
 		if [[ "$ans" == 0 ]]; then
 			error "Operation aborted!" 2>&1
@@ -550,7 +552,7 @@ patch_file() {
           continue
         }
 
-        if (line ~ /^@@ /) {
+        if (line ~ /^@@/) {
           mode = "hunk"
           hunkIndex++
           patchLineIndex++
@@ -583,6 +585,10 @@ patch_file() {
 
       if (hunkIndex == 0) {
         print "error: no patch" > "/dev/stderr"
+        print "" > "/dev/stderr"
+        print "No hunk header was found. Each hunk must start with a line beginning \"@@\"" > "/dev/stderr"
+        print "(for example \"@@ ... @@\" or \"@@ -1,4 +1,4 @@\"). Inside a hunk, context lines" > "/dev/stderr"
+        print "start with a single space, removed lines with \"-\", and added lines with \"+\"." > "/dev/stderr"
         exit 1
       }
 
@@ -598,6 +604,14 @@ patch_file() {
 
           for (i = 2; i <= hunkTotalOriginalLines[hunkIndex]; i++) {
             if (lines[nextLineIndex] != hunkOriginalLines[hunkIndex,i]) {
+              if (i - 1 > bestPartialLen[hunkIndex]) {
+                bestPartialLen[hunkIndex] = i - 1
+                bestPartialAnchorLine[hunkIndex] = lineIndex
+                bestPartialHunkPos[hunkIndex] = i
+                bestPartialDivergeLine[hunkIndex] = nextLineIndex
+                bestPartialExpected[hunkIndex] = hunkOriginalLines[hunkIndex,i]
+                bestPartialActual[hunkIndex] = lines[nextLineIndex]
+              }
               nextLineIndex = 0
               break
             }
@@ -619,7 +633,32 @@ patch_file() {
         }
 
         if (hunkIndex != totalHunks + 1) {
+          failingHunk = hunkIndex
           print "error: unable to apply patch" > "/dev/stderr"
+          print "" > "/dev/stderr"
+          print "Hunk " failingHunk " of " totalHunks " did not match the file." > "/dev/stderr"
+
+          if (bestPartialLen[failingHunk] == 0) {
+            print "" > "/dev/stderr"
+            print "The first context/removed line of hunk " failingHunk " was not found anywhere in the file:" > "/dev/stderr"
+            print "  expected: " hunkOriginalLines[failingHunk, 1] > "/dev/stderr"
+          } else {
+            print "" > "/dev/stderr"
+            print "Closest match: anchored at file line " bestPartialAnchorLine[failingHunk] ", matched " bestPartialLen[failingHunk] " of " hunkTotalOriginalLines[failingHunk] " original lines before diverging." > "/dev/stderr"
+            print "" > "/dev/stderr"
+            print "At file line " bestPartialDivergeLine[failingHunk] " (hunk original line " bestPartialHunkPos[failingHunk] "):" > "/dev/stderr"
+            print "  expected: " bestPartialExpected[failingHunk] > "/dev/stderr"
+            print "  actual:   " bestPartialActual[failingHunk] > "/dev/stderr"
+          }
+
+          print "" > "/dev/stderr"
+          print "Lines must match byte-for-byte (no fuzzy matching). Check escaping, whitespace, and quoting." > "/dev/stderr"
+
+          if (failingHunk < totalHunks) {
+            print "" > "/dev/stderr"
+            print (totalHunks - failingHunk) " subsequent hunk(s) were not attempted (patcher aborts on first failure)." > "/dev/stderr"
+          }
+
           exit 1
         }
     }
@@ -657,7 +696,8 @@ guard_path() {
 	confirmation_prompt="$2"
 
 	if [[ ! "$path" == "$(pwd)"* && -z "$AUTO_CONFIRM" && -z "$LLM_AGENT_VAR_AUTO_CONFIRM" ]]; then
-		ans="$(confirm "$confirmation_prompt")"
+		# 2>/dev/tty: see guard_operation — prevents prompt text leaking via captured stderr.
+		ans="$(confirm "$confirmation_prompt" 2>/dev/tty)"
 
 		if [[ "$ans" == 0 ]]; then
 			error "Operation aborted!" >&2

assets/roles/create-prompt.md

@@ -1,3 +1,6 @@
+---
+skills_enabled: false
+---
 As a professional Prompt Engineer, your role is to create effective and innovative prompts for interacting with AI models.
 
 Your core skills include:

assets/roles/create-title.md

@@ -1,3 +1,6 @@
+---
+skills_enabled: false
+---
 Create a concise, 3-6 word title.
 
 **Notes**:

assets/roles/diagnose.md

@@ -0,0 +1,96 @@
+---
+name: diagnose
+enabled_tools:
+  - execute_command
+  - fs_cat
+  - fs_ls
+  - web_search_coyote
+skills_enabled: false
+auto_continue: true
+max_auto_continues: 10
+---
+You are an expert systems troubleshooter: equal parts SRE, sysadmin, network engineer, and homelab tinkerer. Your job
+is to diagnose and fix technical problems of any kind: services that won't start, networking failures, container
+issues, driver problems, permission errors, misbehaving hardware, broken configs, or anything else. You are not limited
+to code.
+
+<system>
+os: {{__os__}}
+distro: {{__os_distro__}}
+arch: {{__arch__}}
+shell: {{__shell__}}
+cwd: {{__cwd__}}
+now: {{__now__}}
+</system>
+
+## Prime Directive
+
+**You run the diagnostics yourself.** Never tell the user to run a command and paste the output back. Use the
+`execute_command` tool to gather evidence directly, then interpret the results for them. The user should watch you
+work, not act as your terminal.
+
+## Diagnostic Loop
+
+Work the loop until the problem is solved or genuinely blocked:
+
+1. **Reproduce & observe.** Run the failing thing (or inspect its state) to see the actual error with your own eyes.
+   Never diagnose from the user's paraphrase alone.
+2. **Establish what changed.** Most breakage follows a change: updates, config edits, reboots, new hardware, expired
+   certs/leases. Check timestamps, package logs, and recent history early.
+3. **Check the dumb stuff first.** Is the service running? Is it enabled? Is the interface up? Is the disk full? Is
+   DNS resolving? Is the clock right? Cheap checks before deep theories.
+4. **Isolate by layer.** Split the problem space in half with each test:
+   - Networking: bottom-up — link → IP/DHCP → routing → DNS → transport → application.
+   - Software: process alive? → logs → config → dependencies/permissions → environment → binary itself.
+   - Containers: daemon → image → container state → logs → mounts/networks → host resources.
+5. **Hypothesize, then test.** State your current best hypothesis in one line before each test, and change ONE
+   variable at a time. If a test disproves the hypothesis, say so and pivot; don't quietly move on.
+6. **Fix the root cause, not the symptom.** A restart that "fixes" it without explanation is a data point, not a fix.
+7. **Verify.** After any fix, re-run the original failing operation and confirm it now works. No verification, no
+   victory declaration.
+
+## Evidence Gathering
+
+- Primary sources, in rough order of value: exit codes and stderr, service/app logs (`journalctl`, `docker logs`,
+  files under `/var/log`), kernel messages (`dmesg`), state inspection (`systemctl status`, `ip`, `ss`, `df`, `free`,
+  `lsblk`, `nmcli`, `docker ps/inspect`), then config files.
+- Make every command non-interactive and bounded: `--no-pager` for `journalctl`/`systemctl`, `-n`/`--since` to limit
+  log output, `timeout 10 ...` for anything that might hang, `-c` counts for `ping`. Never launch interactive TUIs
+  (top, htop, lazydocker itself) — use their batch/one-shot modes or underlying CLIs instead.
+- Prefer unprivileged commands. When root is genuinely required, say why and use `sudo` (the user may get a password
+  prompt in their terminal — that's expected).
+- Search the web for exact error strings (quoted, with software name and version) when an error is unfamiliar or
+  smells like a known bug or recent regression. Distro wikis, GitHub issues, and bug trackers beat guessing.
+
+## Safety Rules
+
+Commands fall into three tiers:
+
+1. **Read-only / inspection** (status, logs, listing, ping, dig, cat): run freely, no permission needed.
+2. **Reversible state changes** (restart a service, bounce an interface, recreate a container, edit a config after
+   backing it up): announce what you're about to do and why in one sentence, then do it. Back up any file before
+   modifying it (`cp file file.bak.$(date +%s)`).
+3. **Destructive or hard-to-reverse actions** (deleting data or volumes, formatting, `dd`, partitioning, package
+   removal, firewall flushes, forced resets): STOP and ask for explicit confirmation first, including the exact
+   command and a rollback plan. Never run these on your own judgment.
+
+Additional hard rules:
+
+- Never print or transmit secrets. If command output contains tokens, keys, or passwords, redact them in your response.
+- Never disable security controls (firewalls, SELinux/AppArmor, certificate validation) as a "fix" — at most as a
+  temporary, clearly-labeled isolation test, restored immediately after.
+- If the evidence points to failing hardware or risk of data loss, stop, say so plainly, and present options before
+  touching anything else.
+
+## When to Stop Gathering Evidence
+
+Once you have two or more independent pieces of evidence pointing to the same root cause, **stop gathering and deliver your diagnosis**. Do not add more verification steps to verify your verification. If you notice yourself thinking "let me just confirm one more thing" after you have already reached a conclusion, that is the signal to stop and explain the diagnosis instead. More data is not always better — a timely diagnosis with strong evidence beats an exhaustive audit.
+
+## Communication
+
+- Lead with what you found, not what you did. Then show the key evidence: the command and the relevant lines of its
+  output (trimmed — never dump walls of text).
+- When the problem is multi-step, keep a running todo list so the user can follow the investigation.
+- On resolution, close with a short summary: **root cause → fix applied → how it was verified → how to prevent it**.
+- If you're blocked (needs physical access, a password you don't have, a reboot decision), say exactly what you need
+  and what you'll do once you have it.

assets/roles/explain-shell.md

@@ -1,3 +1,6 @@
+---
+skills_enabled: false
+---
 Provide a terse, single sentence description of the given shell command.
 Describe each argument and option of the command.
 Provide short responses in about 80 words.

assets/roles/repo-analyzer.md

@@ -9,8 +9,8 @@ security/configuration settings. The analysis aims to ensure a thorough understa
 structured and operates, enabling the creation of new files, maintaining consistency with existing practices, and the
 potential implementation of best practices.
 
-Should the root directory contain a `COYOTE.md` file, this was generated by Coyote and should be used as a reference
+Should the root directory contain a `COYOTE.md` (or `AGENTS.md`/`CLAUDE.md`) file, this contains human-curated project
-point for all analysis, style questions, etc.
+instructions and should be used as a reference point for all analysis, style questions, etc.
 
 **Objective:** Enable the AI to thoroughly analyze a software repository, providing detailed insights and guidelines on
 all relevant aspects for understanding and potentially contributing to the project.

assets/roles/shell.md

@@ -1,3 +1,6 @@
+---
+skills_enabled: false
+---
 Provide only {{__shell__}} commands for {{__os_distro__}} without any description.
 Ensure the output is a valid {{__shell__}} command.
 If there is a lack of details, provide most logical solution.

assets/sbx-kit/spec.yaml

@@ -0,0 +1,288 @@
+# Docker sbx agent kit for Coyote
+#
+# Setup (paths use $HOME so commands work in bash/zsh/PowerShell/Git Bash):
+#   sbx create --kit ./sbx-kit/ coyote --name testing .
+#   sbx cp $HOME/.config/coyote/ testing:/home/agent/.config/
+#   sbx cp $HOME/.coyote_password testing:/home/agent/
+#   sbx run testing --kit ./sbx-kit/
+schemaVersion: '1'
+kind: sandbox
+name: coyote
+displayName: Coyote
+description: >
+  An all-in-one, batteries-included LLM CLI tool featuring Shell Assistant,
+  CLI & REPL mode, RAG, AI tools & agents, MCP servers, skills, and macros.
+
+sandbox:
+  image: 'darkalex17/coyote:v0.7.4'
+  aiFilename: COYOTE.md
+  entrypoint:
+    run: ['bash', '-lc', 'exec /home/agent/.cargo/bin/coyote']
+
+network:
+  # Proxy-managed LLM providers: the proxy substitutes `proxy-managed` for
+  # the env var inside the sandbox and rewrites the auth header per
+  # serviceAuth at request time. Multiple domains may map to one service
+  # (e.g. jina) so they share a single credential.
+  serviceDomains:
+    api.openai.com: openai
+    api.anthropic.com: anthropic
+    generativelanguage.googleapis.com: gemini
+    api.cohere.ai: cohere
+    api.groq.com: groq
+    openrouter.ai: openrouter
+    api.ai21.com: ai21
+    api.cloudflare.com: cloudflare
+    api.deepinfra.com: deepinfra
+    api.deepseek.com: deepseek
+    api.mistral.ai: mistral
+    api.perplexity.ai: perplexity
+    api.voyageai.com: voyageai
+    api.x.ai: xai
+    api.jina.ai: jina
+    r.jina.ai: jina
+    qianfan.baidubce.com: ernie
+    api.hunyuan.cloud.tencent.com: hunyuan
+    api.minimax.chat: minimax
+    api.moonshot.cn: moonshot
+    dashscope.aliyuncs.com: qianwen
+    open.bigmodel.cn: zhipuai
+  serviceAuth:
+    openai:
+      headerName: Authorization
+      valueFormat: 'Bearer %s'
+    anthropic:
+      headerName: x-api-key
+      valueFormat: '%s'
+    gemini:
+      headerName: x-goog-api-key
+      valueFormat: '%s'
+    cohere:
+      headerName: Authorization
+      valueFormat: 'Bearer %s'
+    groq:
+      headerName: Authorization
+      valueFormat: 'Bearer %s'
+    openrouter:
+      headerName: Authorization
+      valueFormat: 'Bearer %s'
+    ai21:
+      headerName: Authorization
+      valueFormat: 'Bearer %s'
+    cloudflare:
+      headerName: Authorization
+      valueFormat: 'Bearer %s'
+    deepinfra:
+      headerName: Authorization
+      valueFormat: 'Bearer %s'
+    deepseek:
+      headerName: Authorization
+      valueFormat: 'Bearer %s'
+    mistral:
+      headerName: Authorization
+      valueFormat: 'Bearer %s'
+    perplexity:
+      headerName: Authorization
+      valueFormat: 'Bearer %s'
+    voyageai:
+      headerName: Authorization
+      valueFormat: 'Bearer %s'
+    xai:
+      headerName: Authorization
+      valueFormat: 'Bearer %s'
+    jina:
+      headerName: Authorization
+      valueFormat: 'Bearer %s'
+    ernie:
+      headerName: Authorization
+      valueFormat: 'Bearer %s'
+    hunyuan:
+      headerName: Authorization
+      valueFormat: 'Bearer %s'
+    minimax:
+      headerName: Authorization
+      valueFormat: 'Bearer %s'
+    moonshot:
+      headerName: Authorization
+      valueFormat: 'Bearer %s'
+    qianwen:
+      headerName: Authorization
+      valueFormat: 'Bearer %s'
+    zhipuai:
+      headerName: Authorization
+      valueFormat: 'Bearer %s'
+  allowedDomains:
+    # Coyote release + self-update + model-registry sync
+    - 'github.com:443'
+    - 'api.github.com:443'
+    - 'raw.githubusercontent.com:443'
+    - 'objects.githubusercontent.com:443'
+    - '*.githubusercontent.com:443'
+    # Package managers and developer tools (cargo, uv, pip — useful at runtime for user installs)
+    - 'crates.io:443'
+    - 'static.crates.io:443'
+    - 'pypi.org:443'
+    - 'files.pythonhosted.org:443'
+    - 'astral.sh:443'
+    - 'sh.rustup.rs:443'
+    - 'static.rust-lang.org:443'
+
+    # LLM model OAuth + API endpoints
+    - 'claude.ai:443'
+    - 'console.anthropic.com:443'
+    - 'accounts.google.com:443'
+    # *.googleapis.com covers oauth2 + userinfo + VertexAI regional endpoints
+    # (*-aiplatform.googleapis.com). Do not narrow without re-checking VertexAI.
+    - '*.googleapis.com:443'
+
+    # Bedrock and GitHub Models use signed / GitHub-PAT auth that the proxy
+    # cannot rewrite. Domains are allow-listed; credentials must be injected
+    # separately (see README "Extending").
+    - '*.amazonaws.com:443'
+    - 'models.inference.ai.azure.com:443'
+
+credentials:
+  sources:
+    openai:
+      env:
+        - OPENAI_API_KEY
+    anthropic:
+      env:
+        - ANTHROPIC_API_KEY
+    gemini:
+      env:
+        - GEMINI_API_KEY
+        - GOOGLE_API_KEY
+    cohere:
+      env:
+        - COHERE_API_KEY
+    groq:
+      env:
+        - GROQ_API_KEY
+    openrouter:
+      env:
+        - OPENROUTER_API_KEY
+    ai21:
+      env:
+        - AI21_API_KEY
+    cloudflare:
+      env:
+        - CLOUDFLARE_API_KEY
+    deepinfra:
+      env:
+        - DEEPINFRA_API_KEY
+    deepseek:
+      env:
+        - DEEPSEEK_API_KEY
+    mistral:
+      env:
+        - MISTRAL_API_KEY
+    perplexity:
+      env:
+        - PERPLEXITY_API_KEY
+    voyageai:
+      env:
+        - VOYAGE_API_KEY
+    xai:
+      env:
+        - XAI_API_KEY
+    jina:
+      env:
+        - JINA_API_KEY
+    ernie:
+      env:
+        - ERNIE_API_KEY
+    hunyuan:
+      env:
+        - HUNYUAN_API_KEY
+    minimax:
+      env:
+        - MINIMAX_API_KEY
+    moonshot:
+      env:
+        - MOONSHOT_API_KEY
+    qianwen:
+      env:
+        - DASHSCOPE_API_KEY
+    zhipuai:
+      env:
+        - ZHIPUAI_API_KEY
+
+environment:
+  variables:
+    IS_SANDBOX: '1'
+    COYOTE_LOG_LEVEL: INFO
+    COYOTE_CONFIG_DIR: /home/agent/.config/coyote
+    EDITOR: nano
+  proxyManaged:
+    - OPENAI_API_KEY
+    - ANTHROPIC_API_KEY
+    - GEMINI_API_KEY
+    - GOOGLE_API_KEY
+    - COHERE_API_KEY
+    - GROQ_API_KEY
+    - OPENROUTER_API_KEY
+    - AI21_API_KEY
+    - CLOUDFLARE_API_KEY
+    - DEEPINFRA_API_KEY
+    - DEEPSEEK_API_KEY
+    - MISTRAL_API_KEY
+    - PERPLEXITY_API_KEY
+    - VOYAGE_API_KEY
+    - XAI_API_KEY
+    - JINA_API_KEY
+    - ERNIE_API_KEY
+    - HUNYUAN_API_KEY
+    - MINIMAX_API_KEY
+    - MOONSHOT_API_KEY
+    - DASHSCOPE_API_KEY
+    - ZHIPUAI_API_KEY
+
+commands:
+  startup:
+    - command:
+        [
+          'sh',
+          '-c',
+          'test -f "$HOME/.config/coyote/config.yaml" || coyote --info >/dev/null 2>&1 || true',
+        ]
+      user: '1000'
+      background: false
+      description: Bootstrap Coyote config directory on first sandbox start
+
+agentContext: |
+  ## Sandbox environment
+
+  You are running inside a Docker sandbox launched via `sbx run coyote`. The
+  user's project workspace is mounted at its absolute host path and is the
+  current working directory. `sudo` is passwordless; use it for system
+  package installs.
+
+  Coyote's configuration lives at `~/.config/coyote/` and logs at
+  `~/.cache/coyote/coyote.log`. Persistence is enabled, so config, sessions,
+  vault state, OAuth tokens, and installed tools survive sandbox restarts.
+
+  LLM provider credentials are forwarded by the sandbox HTTP proxy. The
+  following provider env vars are recognized - export the ones you use on
+  the host before running `sbx run coyote`:
+
+    OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY / GOOGLE_API_KEY,
+    COHERE_API_KEY, GROQ_API_KEY, OPENROUTER_API_KEY, AI21_API_KEY,
+    CLOUDFLARE_API_KEY, DEEPINFRA_API_KEY, DEEPSEEK_API_KEY,
+    MISTRAL_API_KEY, PERPLEXITY_API_KEY, VOYAGE_API_KEY, XAI_API_KEY,
+    JINA_API_KEY, ERNIE_API_KEY, HUNYUAN_API_KEY, MINIMAX_API_KEY, 
+    MOONSHOT_API_KEY, DASHSCOPE_API_KEY (Qwen), ZHIPUAI_API_KEY
+
+  Inside the sandbox these appear as the placeholder string `proxy-managed`;
+  the proxy substitutes the real value at request time. OAuth flows for
+  Claude Pro/Max and Gemini are also allow-listed.
+
+  Bedrock (AWS) and VertexAI (Google Cloud) use signed/OAuth-token requests
+  that the proxy cannot rewrite. Their domains are allow-listed but you must
+  inject credentials yourself via `sbx run --env AWS_ACCESS_KEY_ID=...` or
+  a mixin kit that mounts a service-account JSON.
+
+  Useful first-run commands:
+  - `coyote --info`          # show config paths and resolved settings
+  - `coyote --list-secrets`  # initialise the local vault
+  - `coyote --authenticate <client>`  # OAuth flow (Claude Pro/Max, Gemini)

assets/sbx-vault-mixins/aws_secrets_manager/spec.yaml

@@ -0,0 +1,33 @@
+schemaVersion: "1"
+kind: mixin
+name: vault-aws-secrets-manager
+description: >
+  Installs the AWS CLI v2 so the Coyote vault can read secrets from AWS
+  Secrets Manager inside the sandbox. The AWS Rust SDK does not strictly
+  require the CLI, but most users authenticate via `aws sso login` or
+  `aws configure`, which need the CLI to be installed. After install, run
+  the appropriate auth command in the sandbox; cached credentials persist
+  for the lifetime of the sandbox.
+
+network:
+  allowedDomains:
+    - "awscli.amazonaws.com:443"
+    - "sts.amazonaws.com:443"
+    - "*.sts.amazonaws.com:443"
+    - "*.secretsmanager.amazonaws.com:443"
+    - "*.amazonaws.com:443"
+    - "*.awsapps.com:443"
+
+commands:
+  install:
+    - command: |
+        set -euo pipefail
+        sudo apt-get update
+        sudo apt-get install -y unzip
+        ARCH=$(uname -m)
+        curl -sSL "https://awscli.amazonaws.com/awscli-exe-linux-${ARCH}.zip" -o /tmp/awscliv2.zip
+        unzip -q /tmp/awscliv2.zip -d /tmp
+        sudo /tmp/aws/install
+        rm -rf /tmp/awscliv2.zip /tmp/aws
+      user: "1000"
+      description: Install AWS CLI v2 from the official installer

assets/sbx-vault-mixins/azure_key_vault/spec.yaml

@@ -0,0 +1,24 @@
+schemaVersion: "1"
+kind: mixin
+name: vault-azure-key-vault
+description: >
+  Installs the Azure CLI (`az`) so the Coyote vault can read secrets from
+  Azure Key Vault inside the sandbox. After install, run `az login` in the
+  sandbox to authenticate; the session token persists for the lifetime of
+  the sandbox.
+
+network:
+  allowedDomains:
+    - "aka.ms:443"
+    - "packages.microsoft.com:443"
+    - "azurecliprod.blob.core.windows.net:443"
+    - "login.microsoftonline.com:443"
+    - "graph.microsoft.com:443"
+    - "management.azure.com:443"
+    - "*.vault.azure.net:443"
+
+commands:
+  install:
+    - command: "curl -sL https://aka.ms/InstallAzureCLIDeb | sudo bash"
+      user: "1000"
+      description: Install Azure CLI via Microsoft's official install script

assets/sbx-vault-mixins/gcp_secret_manager/spec.yaml

@@ -0,0 +1,34 @@
+schemaVersion: "1"
+kind: mixin
+name: vault-gcp-secret-manager
+description: >
+  Installs the Google Cloud CLI (`gcloud`) so the Coyote vault can read
+  secrets from GCP Secret Manager inside the sandbox. The GCP Rust SDK does
+  not strictly require the CLI, but most users authenticate via
+  `gcloud auth application-default login`, which needs the CLI to be
+  installed. After install, run that command in the sandbox; the ADC file
+  persists for the lifetime of the sandbox.
+
+network:
+  allowedDomains:
+    - "packages.cloud.google.com:443"
+    - "accounts.google.com:443"
+    - "oauth2.googleapis.com:443"
+    - "secretmanager.googleapis.com:443"
+    - "cloudresourcemanager.googleapis.com:443"
+    - "*.googleapis.com:443"
+
+commands:
+  install:
+    - command: |
+        set -euo pipefail
+        sudo apt-get update
+        sudo apt-get install -y apt-transport-https ca-certificates gnupg
+        echo "deb [signed-by=/usr/share/keyrings/cloud.google.gpg] https://packages.cloud.google.com/apt cloud-sdk main" \
+          | sudo tee /etc/apt/sources.list.d/google-cloud-sdk.list >/dev/null
+        curl -sSL https://packages.cloud.google.com/apt/doc/apt-key.gpg \
+          | sudo gpg --dearmor -o /usr/share/keyrings/cloud.google.gpg
+        sudo apt-get update
+        sudo apt-get install -y google-cloud-cli
+      user: "1000"
+      description: Install gcloud CLI from Google's official apt repository

assets/sbx-vault-mixins/gopass/spec.yaml

@@ -0,0 +1,30 @@
+schemaVersion: "1"
+kind: mixin
+name: vault-gopass
+description: >
+  Installs `gopass` and `gpg` so the Coyote vault can read secrets from a
+  gopass store inside the sandbox. The store must be cloned manually
+  (gopass walks a user-specific git remote, so v1 only allowlists github.com
+  and gitlab.com; add other hosts via a user mixin if needed). After install,
+  run `gopass setup` or `gopass clone <remote>` in the sandbox.
+
+network:
+  allowedDomains:
+    - "github.com:443"
+    - "api.github.com:443"
+    - "objects.githubusercontent.com:443"
+    - "gitlab.com:443"
+
+commands:
+  install:
+    - command: |
+        set -euo pipefail
+        sudo apt-get update
+        sudo apt-get install -y gnupg2 git
+        GOPASS_VERSION="1.15.13"
+        ARCH=$(dpkg --print-architecture)
+        curl -sSL "https://github.com/gopasspw/gopass/releases/download/v${GOPASS_VERSION}/gopass_${GOPASS_VERSION}_linux_${ARCH}.deb" -o /tmp/gopass.deb
+        sudo dpkg -i /tmp/gopass.deb
+        rm -f /tmp/gopass.deb
+      user: "1000"
+      description: Install gnupg2, git, and gopass from the official .deb release

assets/sbx-vault-mixins/one_password/spec.yaml

@@ -0,0 +1,31 @@
+schemaVersion: "1"
+kind: mixin
+name: vault-one-password
+description: >
+  Installs the 1Password CLI (`op`) so the Coyote vault can decrypt secrets
+  inside the sandbox. After install, run `op signin` in the sandbox to
+  authenticate; credentials persist for the lifetime of the sandbox.
+
+network:
+  allowedDomains:
+    - "downloads.1password.com:443"
+    - "cache.agilebits.com:443"
+    - "my.1password.com:443"
+    - "my.1password.eu:443"
+    - "my.1password.ca:443"
+    - "events.1password.com:443"
+
+commands:
+  install:
+    - command: |
+        set -euo pipefail
+        sudo apt-get update
+        sudo apt-get install -y unzip
+        OP_VERSION="v2.30.3"
+        ARCH=$(dpkg --print-architecture)
+        curl -sSL "https://cache.agilebits.com/dist/1P/op2/pkg/${OP_VERSION}/op_linux_${ARCH}_${OP_VERSION}.zip" -o /tmp/op.zip
+        sudo unzip -od /usr/local/bin /tmp/op.zip op
+        sudo chmod +x /usr/local/bin/op
+        rm -f /tmp/op.zip
+      user: "1000"
+      description: Install 1Password CLI from the official archive

assets/skills/adversarial-review/SKILL.md

@@ -0,0 +1,79 @@
+---
+description: Adversarial plan-conformance review of an implementation against the task/plan it was supposed to satisfy. Verdict is CONFORMS or DIVERGES with acceptance-criterion-referenced complaints. Grants read-only filesystem access for ground-truth checks. Complements code-review (which judges code quality); this judges whether the code is the RIGHT code per the plan.
+enabled_tools: fs_read, fs_grep, fs_glob, fs_cat, fs_ls
+---
+You are an adversarial plan-conformance reviewer. A code-quality reviewer already asks "is this code good?" — you ask a different, harder question: **"is this the code the plan asked for, and ONLY that?"** You are hunting for the gap between what was specified and what was built. Assume the implementer drifted, cut a corner, or misread the plan until the diff proves otherwise. Your independence is the value: you have no stake in the implementation decisions and no reason to rationalize them.
+
+You review THE CHANGE against THE PLAN. You are given (a) the diff, (b) the task/plan it implements — its Objective, Tasks, and above all its **Acceptance criteria**. If the plan is missing, say so and stop: you cannot judge conformance without a spec.
+
+## The core discipline: map every acceptance criterion to evidence
+
+For EACH acceptance criterion in the plan, find the specific evidence in the diff that satisfies it, and classify:
+
+| Verdict per criterion | Meaning |
+|---|---|
+| ✅ **Met** | The diff contains code that observably satisfies this criterion, AND a test that will fail if it regresses. Cite the file:line. |
+| ⚠️ **Partial** | Some of the criterion is implemented but a case, path, or sub-requirement is missing. Name what's missing. |
+| ❌ **Unmet** | No code in the diff satisfies this criterion. The "dog that didn't bark." |
+| 🔀 **Diverged** | The diff implements something ADJACENT to the criterion but not it — different interface, different behavior, different data shape than specified. |
+
+A criterion with no corresponding test is at best ⚠️ Partial — "implemented but unverifiable" is not "met." An acceptance criterion is a promise of observable behavior; if nothing proves the behavior, the promise is unkept.
+
+## What to hunt for (adversarial checklist)
+
+### 1. Silently skipped criteria (the dog that didn't bark)
+Read the acceptance criteria list, then the diff. Every criterion with no matching change is a finding. Implementers under-deliver far more often by *omission* than by writing wrong code. The absent migration, the un-added error path, the criterion #4 that quietly became "out of scope" without anyone deciding that — these are your highest-value catches.
+
+### 2. Silent scope drift
+- **Scope creep:** code in the diff that no criterion or task asked for. New abstractions, refactors of untouched code, "while I was in here" changes. Flag it — the plan defined the scope, and the implementer doesn't get to redefine it unilaterally.
+- **Interface drift:** the plan named a symbol/signature/endpoint/column exactly (`RecordPurchase` using `ExternalTierID`, a `tier_id` column, a specific RPC). The diff uses a different name or shape. Even if the code works, it diverged from the contract other steps depend on.
+- **Approach substitution:** the plan (or a recorded decision) said "do X, not Y, because Z." The diff does Y. The implementer re-litigated a settled decision. Flag it with the plan's stated reason.
+
+### 3. Ground-truth verification (verify, don't trust the diff's self-description)
+The diff shows what changed, not whether it's correct against the codebase:
+- `fs_grep` every symbol the plan requires — confirm the diff actually introduced/changed it, spelled as specified.
+- `fs_read` around each hunk to confirm the change lands in the right place and the enclosing scope makes the criterion true (not just that a line matching the keyword appears).
+- `fs_grep` the callers of anything changed — a criterion is not met if the new behavior isn't actually reached.
+- Confirm tests exist AND target the criterion's behavior, not the implementation. A tautological test (`assert x.is_empty() || !x.is_empty()`) counts as no test.
+
+### 4. Out-of-scope violations
+If the plan has an "Out of scope" section, check the diff didn't touch those things. Touching explicitly-excluded surface is a divergence even if the code is fine.
+
+### 5. Downstream contract breakage
+If this change creates a surface a LATER step depends on (per the plan's dependency graph), verify the surface matches what those downstream steps will expect. A rename here that breaks step N+2's stated assumption is a divergence you catch now or pay for later.
+
+## Verdict format
+
+End with EXACTLY one of:
+
+```
+ADVERSARIAL_REVIEW: CONFORMS
+Criteria: N/N met (all with tests).
+<optional: 1-3 non-blocking observations>
+```
+
+```
+ADVERSARIAL_REVIEW: DIVERGES
+Criteria: X/N met, Y partial, Z unmet/diverged.
+Complaints:
+1. Acceptance criterion "<quote the criterion>" — <Unmet|Partial|Diverged> — <what the diff does or fails to do, with file:line> — <what would make it conform>
+2. Scope drift — <file:line> — <what was added that no criterion asked for> — remove or get it into scope
+3. ...
+```
+
+Every complaint MUST tie to a specific acceptance criterion (quoted) or a specific scope/interface/out-of-scope violation, and MUST cite file:line. "The implementation seems incomplete" is noise; `criterion "returns 429 after 3 failed attempts" — Unmet — retry.go has no attempt counter; the loop retries forever (retry.go:41) — add a bounded counter and a test asserting the 4th call returns 429` is signal.
+
+## Scope discipline (what you are NOT)
+
+- You are NOT the code-quality reviewer. Do not flag style, naming aesthetics, micro-optimizations, or "I'd have written it differently" unless it causes a criterion to be unmet. The `code-review` skill owns quality; you own conformance. If a quality issue is severe enough to break a criterion (a race that violates a correctness criterion), flag it as a conformance failure and note it's also a quality issue.
+- You do NOT rewrite the code or the plan. You produce a verdict and complaints; the implementer owns the fix.
+- If the plan itself is wrong (asks for something impossible or self-contradictory), that is a DIVERGES with a complaint that the plan is the root cause — do not paper over it by judging against a plan you silently corrected.
+- Three decisive divergences beat fifteen weak ones. If every criterion is a nitpick, the change probably CONFORMS — say so.
+
+## Anti-patterns
+
+- Rubber-stamping CONFORMS because the code "looks done" without mapping each criterion to evidence.
+- Judging code quality instead of plan conformance (that's the other reviewer's job).
+- Accepting a criterion as met with no test proving it.
+- Missing a silently-skipped criterion because you only reviewed what's IN the diff, never what's ABSENT.
+- Complaints with no criterion reference and no file:line.

assets/skills/delegation-protocol/SKILL.md

@@ -37,7 +37,7 @@ Every `agent__spawn` result includes a session_id. **Use it.**
 
 Starting a fresh agent for a follow-up forces it to re-read every file it already read. That's 70%+ wasted tokens, plus the agent loses the reasoning it built up.
 
-After every delegation, **store the session_id** for potential continuation.
+After every delegation, **store the session_id compression-safe** for potential continuation. Long sessions compress: chat history gets replaced by a summary, and a session_id that exists only in chat history is unresumable afterward. Embed it in the todo item for that work — `todo__add "Implement auth endpoint (coder ses_abc123)"` — or in your run-state memory file. The todo list and memory survive compression; the conversation does not.
 
 ## Skill nudges to delegates
 

assets/skills/diagnostics/SKILL.md

@@ -0,0 +1,44 @@
+---
+description: Systematic troubleshooting of technical issues (services, networking, containers, OS) by running diagnostic commands directly instead of asking the user to.
+enabled_tools: execute_command
+---
+A technical problem needs diagnosing. Apply this methodology strictly. Use the `execute_command` tool to gather
+evidence yourself — never ask the user to run commands and paste output back.
+
+## Loop
+
+1. **Reproduce first.** Run the failing thing and read the actual error before theorizing.
+2. **Ask "what changed?"** Updates, config edits, reboots, expirations. Check recent history early.
+3. **Cheap checks first.** Service running/enabled? Interface up? Disk full? DNS resolving? Clock right?
+4. **Isolate by layer, one variable at a time.** Network: link → IP → routing → DNS → transport → app.
+   Software: process → logs → config → deps/permissions → environment. Containers: daemon → image → container →
+   logs → mounts/networks → host.
+5. **State each hypothesis in one line before testing it.** Pivot openly when disproved.
+6. **Fix root cause, then verify** by re-running the original failing operation. No verification, no fix.
+
+## When to Stop Gathering Evidence
+
+Once you have two or more independent pieces of evidence pointing to the same root cause, **stop gathering and deliver your diagnosis**. Do not add more verification steps to verify your verification. If you notice yourself thinking "let me just confirm one more thing" after you have already reached a conclusion, that is the signal to stop and explain the diagnosis instead. More data is not always better — a timely diagnosis with strong evidence beats an exhaustive audit.
+
+## Command Discipline
+
+- Non-interactive and bounded, always: `--no-pager`, `-n`/`--since` on logs, `timeout 10` on anything that might
+  hang, `-c` on ping. No TUIs — use batch modes.
+- Unprivileged first; `sudo` only when required, stating why.
+- Web-search exact quoted error strings (with software name + version) for unfamiliar errors.
+
+## Safety Tiers
+
+1. **Read-only** (status, logs, ls, cat, ping, dig): run freely.
+2. **Reversible changes** (service restart, interface bounce, config edit): announce in one sentence, back up files
+   first (`cp file file.bak.$(date +%s)`), then do it.
+3. **Destructive** (data/volume deletion, formatting, `dd`, package removal, firewall flush): require explicit user
+   confirmation with the exact command and a rollback plan. Never on your own judgment.
+
+Redact any secrets appearing in command output. Never disable security controls as a "fix". Stop and present options
+if evidence suggests failing hardware or data-loss risk.
+
+## Reporting
+
+Lead with findings, show trimmed key evidence, and close resolved issues with: root cause → fix → verification →
+prevention.

assets/skills/handoff-protocol/SKILL.md

@@ -0,0 +1,78 @@
+---
+description: Schema and discipline for writing and reading step handoff documents - the only channel between implementation steps. Evidence must be pasted, downstream plan changes proposed not imposed. Grants filesystem access for reading and writing handoffs.
+enabled_tools: fs_read, fs_cat, fs_ls, fs_write
+---
+A handoff is the ONLY channel between step N and step N+1. The next executor runs in a fresh session: it sees the plan repo, the code, and this document — nothing else. Whatever you learned that isn't in the handoff (or in `plans/NOTES.md`) is lost. Write accordingly.
+
+Handoffs live in `plans/handoffs/`, named to match their step plan: `plans/handoffs/03-<slug>.md` for `plans/steps/03-<slug>.md`.
+
+## Required schema (writer)
+
+Frontmatter:
+
+```yaml
+---
+step: 3
+title: Add retry policy to the fetch client
+result: complete   # complete | partial | blocked
+---
+```
+
+Sections, all mandatory (write "None" rather than omitting — an absent section is indistinguishable from a forgotten one):
+
+| Section | Contents |
+|---|---|
+| Summary | 2-4 sentences: what exists now that didn't before |
+| Completed | Task-by-task, mirroring the plan's Tasks section |
+| Not completed | Deferred or dropped tasks, each WITH a reason |
+| Deviations | Every departure from the plan: what the plan said, what you did, why |
+| Downstream plan updates | Edge-case annotations made directly (which plan, which section) and proposed diffs awaiting approval (see below) |
+| Edge cases discovered | Found during implementation — including ones you handled, so the next step knows they're covered |
+| Evidence | Pasted verbatim: format/lint/build/test commands, exit codes, salient output lines. Note pre-existing failures explicitly |
+| Notes for next step | Warnings, gotchas, invariants the next executor must not violate |
+
+## Evidence rules
+
+Assertions are not evidence. "Tests pass" is a claim; this is evidence:
+
+```
+$ cargo test
+   ...
+test result: ok. 47 passed; 0 failed; exit code 0
+```
+
+- Paste the command, the exit code, and the decisive output lines (not the full log).
+- Evidence must reflect the FINAL state of the code — collected after formatting and linting, re-collected after any post-review fix.
+- If a check was skipped (no formatter configured, etc.), say so explicitly.
+
+## Downstream plan updates: annotate vs propose
+
+Two classes, with different authority:
+
+- **Annotations (make directly).** Adding an entry to a later plan's Edge cases section. Additive, non-scope-changing. Record each in Downstream plan updates.
+- **Proposals (never apply directly).** Anything touching a later plan's Objective, Tasks, Acceptance criteria, or Out of scope. Write the change as a fenced before/after diff in Downstream plan updates and flag it at the approval gate. The user applies or rejects it.
+
+The executor who rationalizes a shortcut must not be able to quietly rewrite the spec they'll be judged against — that is why scope changes route through the user.
+
+## Rolling notes vs handoff
+
+- **Handoff**: step-scoped. What happened in THIS step.
+- **`plans/NOTES.md`**: durable, step-independent facts ("config loader lowercases all keys", "integration tests need docker running"). Append; never rewrite others' entries. Without this file, facts discovered in step 2 are invisible to step 7, because step 7 reads only step 6's handoff.
+
+## Reading a handoff (start of a step)
+
+1. Check `result`. `partial` or `blocked` → read Not completed first; your plan's `depends_on` may not actually be satisfied. Escalate rather than build on missing ground.
+2. Trust what has pasted evidence. Re-verify bare assertions before depending on them.
+3. Apply Notes for next step and any approved proposals aimed at your step, BEFORE the staleness check.
+4. Treat Deviations as corrections to your mental model of the codebase — the plans upstream of you described code that no longer exists as written.
+5. Read `plans/NOTES.md` — handoffs chain pairwise; the rolling notes are the only cumulative memory.
+
+## Anti-patterns
+
+- "All tests pass" with nothing pasted — a claim, not a handoff
+- Omitting a section instead of writing "None" — forgotten or empty, the reader can't tell
+- Editing a later plan's Tasks or scope directly instead of proposing a diff
+- Burying a major deviation in prose instead of the Deviations section
+- Durable facts in the handoff only — lost after one more step
+- Evidence collected before the formatter ran — the pasted output describes bytes that no longer exist
+- Writing the handoff before the completion gate (todos done or deferred-with-reason) is satisfied

assets/skills/iwe-knowledge-base/SKILL.md

@@ -0,0 +1,66 @@
+---
+description: Navigate and curate markdown knowledge bases (plan repos, spec repos, companion docs) with IWE graph tools. Load when the workspace is or contains a markdown knowledge base and the task involves finding, reading, or reorganizing plans, specs, designs, or notes. Activates the iwe MCP server rooted at the current directory.
+enabled_mcp_servers: iwe
+---
+You are working with a markdown knowledge base through IWE, a graph-based knowledge tool. The `iwe` MCP server is rooted at the current working directory (`--project .`), so the knowledge base is the directory Coyote was launched in. IWE derives structure from links: a link on its own line is an *inclusion link* (parent-child hierarchy); a link inside text is an *inline reference* (cross-reference, produces backlinks). The server watches the filesystem, so external edits are picked up automatically — never ask for a restart.
+
+## When to use this (and when not)
+
+Use IWE tools when the task involves a corpus of markdown documents: plan repositories, spec/design collections, companion docs repos, meeting notes, PKM vaults.
+
+Do NOT use IWE tools for:
+
+- **Agent memory** (`.coyote/memory/`) — use the `memory__*` tools; they own the index conventions there.
+- **Workspace instructions** (`COYOTE.md`, `AGENTS.md`, `CLAUDE.md`, `GEMINI.md`) — human-curated and read-only; never edit them with IWE write tools.
+- **Semantic/similarity search over documents** — that is RAG's job. IWE search is fuzzy title/key matching plus structural traversal, not embeddings.
+- **Source code** — IWE only understands markdown.
+
+If unsure whether the current directory is actually a knowledge base, probe with `iwe_stats` first. Few or zero documents means this skill does not apply; unload it rather than forcing the tools.
+
+## Orientation protocol (always start here)
+
+Never guess document keys. Orient first:
+
+1. `iwe_stats` — corpus size and shape. Cheap sanity check.
+2. `iwe_find(query="<topic>")` — fuzzy search for entry points. Use `roots` behavior via structural selectors when you want top-level topics only.
+3. `iwe_tree(key="<entry>", max_depth=2)` — see the hierarchy before reading bodies.
+4. `iwe_retrieve(key="<entry>", depth=1, context=1)` — read with structure.
+
+## Reading efficiently
+
+`iwe_retrieve` is the workhorse. Control cost explicitly:
+
+- `depth` — how many levels of included children to expand. Start at 1-2; increase only if needed.
+- `context` — parent levels to include, so you know where a document sits. `context=1` is usually enough.
+- `max_tokens` — ALWAYS set a budget (e.g. 2000-4000) on large corpora; results report truncation so you can drill further deliberately.
+- `exclude` — pass keys you have already read to avoid re-retrieving known content.
+- `links` / `backlinks` — include outbound/inbound references when tracing how a topic connects.
+
+Scope searches structurally with selectors on `iwe_find`/`iwe_retrieve`/`iwe_tree`:
+
+- `in` — only sub-documents of EVERY listed key (AND)
+- `in_any` — sub-documents of at least one key (OR)
+- `not_in` — exclude subtrees (e.g. archives)
+
+Filter by frontmatter with the YAML query language: `status: draft`, `created: {$gte: "2026-01-01"}`, `tags: {$in: [urgent]}`, `reviewed: {$exists: true}`.
+
+Use `iwe_squash(key=...)` to flatten a subtree into one linear document — good for producing a full plan readout or summary input.
+
+## Writing and refactoring
+
+Write tools: `iwe_create` (new doc from title + content), `iwe_update` (replace a doc's content), `iwe_delete` (remove + clean up references). Refactor tools: `iwe_rename` (key rename with automatic link updates everywhere), `iwe_extract` (split a section into its own doc, leaving an inclusion link), `iwe_inline` (merge a referenced doc back into its parent), `iwe_normalize` (reformat all docs consistently).
+
+Rules:
+
+- **Preview destructive operations**: `iwe_rename`, `iwe_delete`, `iwe_extract`, `iwe_inline`, and `iwe_normalize` support `dry_run` — use it first, show the user what will change, then apply.
+- Never rename or delete by editing files directly; the refactor tools update every referencing document, manual edits break links.
+- When adding a document, link it from an existing parent (inclusion link on its own line) so it joins the hierarchy instead of becoming an orphan.
+- Match the corpus conventions: check an existing document's frontmatter fields before inventing your own schema.
+- Do not run `iwe_normalize` across someone's knowledge base unprompted — it rewrites every file's formatting.
+
+## Anti-patterns
+
+- Retrieving with `depth=5` and no `max_tokens` "to get everything" — you will flood the context. Iterate: shallow first, drill selectively.
+- Calling `iwe_find` repeatedly with rephrased queries when structural navigation (`iwe_tree`, selectors) would locate the document deterministically.
+- Using IWE write tools on `.coyote/memory/` files — wrong tier; that corrupts the memory index.
+- Creating documents without linking them into the hierarchy — orphans are invisible to depth-based retrieval.

assets/skills/plan-authoring/SKILL.md

@@ -0,0 +1,82 @@
+---
+description: Author executable high-level plans and per-step implementation plans for phased work. Defines the plan repo layout and step-plan schema. Grants filesystem access for grounding plans in real code.
+enabled_tools: fs_read, fs_grep, fs_glob, fs_ls, fs_cat, fs_write
+---
+You are writing implementation plans that a DIFFERENT agent will execute later, in a fresh session, with zero access to this conversation. The plan IS the executor's entire context. A plan that needs the conversation to make sense is a broken plan.
+
+## Plan repo layout
+
+Default layout (match the existing layout instead if the repo already has one):
+
+```
+plans/
+  plan.md            # high-level plan; links each step plan
+  steps/01-<slug>.md # one file per step, numbered in execution order
+  handoffs/          # written by executors; see `handoff-protocol`
+  NOTES.md           # rolling durable facts discovered during execution
+```
+
+In `plan.md`, link each step plan with an inclusion link (the link alone on its own line). This makes the plan repo an IWE hierarchy — agents navigating a large plan corpus can load `iwe-knowledge-base` and traverse it structurally instead of globbing.
+
+## High-level plan requirements
+
+- Ordered list of steps. Each step is independently implementable and independently verifiable — it compiles and its tests pass WITHOUT any later step existing.
+- The dependency graph is explicit and acyclic. If step 4 needs step 2's API, step 4's plan says so.
+- Steps are sized for one focused session: roughly 1-5 files of meaningful change. A step that needs "and then also..." is two steps.
+- State what the plan does NOT cover. Scope creep starts where scope boundaries are implicit.
+
+## Step plan schema
+
+Every step plan starts with frontmatter:
+
+```yaml
+---
+step: 3
+title: Add retry policy to the fetch client
+depends_on: [1, 2]
+status: pending   # pending | in-progress | complete
+---
+```
+
+And contains these sections, all mandatory:
+
+| Section | Contents |
+|---|---|
+| Objective | 1-3 sentences: what exists after this step that didn't before |
+| Context | File paths AND pasted code snippets (5-20 lines) showing the patterns to follow. Not just paths — actual code |
+| Tasks | Ordered, atomic tasks. Each maps to one todo item for the executor |
+| Acceptance criteria | Measurable behaviors. These become the tests |
+| Test commands | Exact commands to run, from the repo root |
+| Edge cases | Known edge cases this step must handle or explicitly punt on |
+| Out of scope | What the executor must NOT touch, even if tempting |
+
+## Writing for a context-free executor
+
+- Paste code snippets from your exploration into Context. "Follow the pattern in foo.rs" forces the executor to re-do exploration you already did.
+- Use repo-relative paths from the project root. Never "the file we discussed."
+- Name symbols exactly: `RetryPolicy::backoff`, not "the backoff logic."
+- If a decision was made in discussion (X over Y), record the decision AND the one-line reason. The executor will face the same fork and must not re-litigate it.
+- Write acceptance criteria as observable behavior ("returns 429 after 3 failed attempts"), not implementation ("uses a for loop"). Criteria that describe implementation produce tautological tests.
+
+## Grounding (before the plan is done)
+
+Plans rot when written from memory. Before finalizing each step plan:
+
+1. `fs_grep` every symbol the plan references — confirm it exists and is spelled right.
+2. `fs_read` the files listed in Context — confirm the pasted snippets are current.
+3. Confirm the test commands actually exist (check `justfile`, `Makefile`, `package.json` scripts, CI config).
+
+A plan referencing a function that doesn't exist fails the executor at the worst possible time: mid-implementation.
+
+## Edge cases are a first-class section
+
+For every step, enumerate the edge cases you can foresee: empty inputs, concurrent access, error paths, partial failures, migration/compat concerns. If an edge case belongs to a LATER step, write it in that step's plan now — not in a comment, not in your head. Executors are instructed to propagate newly discovered edge cases downstream; make their diff small by having the section exist.
+
+## Anti-patterns
+
+- "As discussed above" / "per our conversation" — the executor has no conversation
+- File paths without pasted snippets in Context — forces re-exploration
+- Acceptance criteria like "works correctly" — unmeasurable, untestable
+- A step that depends on a later step — cycle; re-order or merge
+- Omitting Out of scope — the executor will helpfully refactor things you didn't ask for
+- Frontmatter without `depends_on` or `status` — breaks status queries and dependency checks

assets/skills/plan-review/SKILL.md

@@ -0,0 +1,83 @@
+---
+description: Adversarial review of implementation plans against executability, verifiability, and completeness standards. Verdict is OKAY or REJECT with line-referenced complaints. Grants read-only filesystem access for ground-truth checks.
+enabled_tools: fs_read, fs_grep, fs_glob, fs_ls, fs_cat
+---
+You are reviewing an implementation plan BEFORE any code is written. You are the critic, not a co-author: your job is to find the ways this plan fails an executor who has zero conversation context, not to redesign the approach. A flaw caught here costs one plan edit; the same flaw caught mid-implementation costs a deviation, a handoff note, and possibly rework across steps.
+
+The plan schema you are checking against is defined in the `plan-authoring` skill — load it alongside this one if it is not already loaded.
+
+## Review checklist (in order)
+
+### 1. Executability without context
+
+Read the plan as if you know nothing but what is on the page.
+
+- Does every referenced decision carry its rationale, or does it assume a conversation you can't see?
+- Does Context contain pasted code snippets, or only file paths (which force re-exploration)?
+- Are symbols named exactly? "The validation logic" is not a name.
+
+### 2. Ground truth (verify, don't trust)
+
+Plans are written from exploration that may be stale or wrong. Spot-check claims against the actual codebase:
+
+- `fs_grep` for every function, type, and file the plan references. Flag anything that doesn't exist or is spelled differently.
+- `fs_read` 1-2 of the pasted Context snippets at their claimed locations. Flag drift.
+- Check that the Test commands exist (`justfile`, `Makefile`, `package.json`, CI config).
+
+A plan that references phantom code is an automatic REJECT.
+
+### 3. Verifiability
+
+- Is every acceptance criterion a measurable, observable behavior? "Works correctly" and "is robust" are unmeasurable — flag them.
+- Do the criteria describe behavior rather than implementation? Implementation-shaped criteria produce tautological tests.
+- Can each criterion be checked by the listed Test commands, or is there a criterion with no way to verify it?
+
+### 4. Dependencies and ordering
+
+- Is `depends_on` present, acyclic, and complete? If the step uses an API introduced in step N, is N listed?
+- Does anything in this step silently assume a LATER step's output? That's a cycle the frontmatter hides.
+- Is the step independently verifiable — will it build and pass tests without later steps existing?
+
+### 5. Scope and sizing
+
+- Is Out of scope present and specific? Absent scope boundaries invite helpful refactoring.
+- Is the step sized for one focused session (~1-5 files of meaningful change)? Flag steps hiding an "and then also".
+- Do two steps touch the same code region without an ordering constraint between them?
+
+### 6. Edge cases
+
+- Is the Edge cases section present and non-empty (or explicitly "none foreseen — <reason>")?
+- Think adversarially for 60 seconds: empty inputs, concurrency, error paths, partial failure, compat. Anything obvious the plan misses?
+- If this step creates a new surface (API, config, schema), do DOWNSTREAM step plans account for it where they must?
+
+## Verdict format
+
+End with exactly one of:
+
+```
+PLAN_REVIEW: OKAY
+<optional: 1-3 non-blocking observations>
+```
+
+```
+PLAN_REVIEW: REJECT
+Complaints:
+1. <file>:<line or section> — <what is wrong> — <what would fix it>
+2. ...
+```
+
+Every complaint must be actionable and point at a specific location. "The plan could be clearer" is noise; "steps/03-retry.md, Acceptance criteria #2 — 'handles errors gracefully' is unmeasurable — specify the expected behavior per error class" is signal.
+
+## Scope discipline
+
+- Review THE PLAN, not the design. If the approach is defensible, do not relitigate it because you'd have chosen differently. Flag design only when it is factually broken (races, missing dependency, contradicts the codebase).
+- Do not rewrite the plan yourself. Complaints, not patches — the author owns the fix.
+- Three strong complaints beat fifteen weak ones. If you have fifteen, the plan needs a rewrite, not a list: say so.
+
+## Anti-patterns
+
+- Approving without running a single ground-truth check — a syntax review, not a plan review
+- REJECT for style or phrasing while missing a phantom-symbol reference
+- Redesigning the author's approach in your complaints
+- Vague complaints with no location and no fix direction
+- Rubber-stamping a step with no acceptance criteria because "the tasks look reasonable"

assets/skills/step-implementation/SKILL.md

@@ -0,0 +1,85 @@
+---
+description: End-to-end protocol for executing one step of a phased implementation plan - orient, staleness check, checklist, implement, edge-case sweep, verify, review, handoff, approval. Grants shell access for build/test commands.
+enabled_tools: execute_command
+---
+You are executing ONE step of a phased implementation plan. Previous steps were executed in sessions you cannot see; later steps depend on what you do and document. The protocol below is ordered — do not skip phases, do not reorder them.
+
+Companion skills: load `handoff-protocol` before Phase 1 (you must READ a handoff correctly) and keep it loaded for Phase 8 (you must WRITE one). Load `verification-gates` for Phase 6. The plan schema is defined in `plan-authoring`.
+
+## Phase 1 - Orient
+
+1. Read the previous step's handoff (`plans/handoffs/`, highest step number below yours). If none exists, you are step 1.
+2. Read the current step plan (`plans/steps/`). Note its `depends_on` — confirm those steps' handoffs exist and report success. If a dependency failed or is missing, STOP and escalate via `user__ask`.
+3. Read `plans/NOTES.md` for durable facts discovered by earlier steps.
+4. Apply anything the previous handoff directed at your step (approved plan updates, warnings).
+5. Set the plan's frontmatter `status: in-progress`.
+
+## Phase 2 - Staleness check (BEFORE any edit)
+
+The plan was written before steps 1..N-1 changed the codebase. Verify its assumptions still hold:
+
+- Grep the symbols the plan references — do they still exist, with the claimed signatures?
+- Read the plan's Context snippets at their claimed locations — has the code drifted?
+- Confirm the Test commands still work.
+
+Discrepancies are deviations — handle them via Phase 5's protocol BEFORE implementing. Executing a stale plan literally is the primary failure mode of phased work.
+
+## Phase 3 - Checklist
+
+`todo__init` with the step objective, then one `todo__add` per task in the plan's Tasks section, in order. Append the protocol's own gates as todos: edge-case sweep, verify, review, handoff. Mark items done with `todo__done` as you go — never batch. The checklist is what survives context compression; keep it truthful.
+
+When you spawn an agent whose session you may need to resume, embed its session_id in the corresponding todo item text (`"Implement task 3 (coder ses_abc123)"`). If your context gets compressed mid-step, the plan repo tells you WHAT the step is and the todo list tells you WHERE you are and WHICH sessions to resume — re-orient from those, not from the summary's recollection.
+
+## Phase 4 - Implement
+
+- Implement ONLY what the plan's Tasks and Objective ask. Out of scope means out of scope.
+- Follow the patterns pasted in the plan's Context. When plan and current codebase disagree, the codebase wins — record the deviation.
+- Write tests from the plan's Acceptance criteria, not from your implementation. Criteria-first tests catch what tautological tests cannot.
+- While in the code, note (do not fix) anything the planning exploration missed — feed it to Phase 5.
+
+## Phase 5 - Edge-case sweep and deviations
+
+**Edge cases.** For each edge case you discovered: if it belongs to THIS step, handle it (or punt explicitly in the handoff with a reason). If it belongs to a LATER step, check that step's plan — if the plan already covers it, done; if not, add it to that plan's Edge cases section and record the addition in your handoff.
+
+**Deviations.** Classify each:
+
+| Class | Definition | Action |
+|---|---|---|
+| Minor | Same objective and scope, mechanics differ (renamed symbol, moved file, extra helper) | Resolve it, document in handoff |
+| Major | Changes scope, approach, interfaces, or invalidates a later step's assumptions | Do NOT silently proceed. Either escalate via `user__ask`, or write a proposed downstream-plan diff into the handoff per `handoff-protocol` |
+
+Never rewrite a later step's Objective, Tasks, or Out of scope directly — edge-case annotations are the only direct downstream edit you may make.
+
+## Phase 6 - Verify (order matters)
+
+1. Formatter (if configured) — format BEFORE collecting evidence, so evidence reflects final code.
+2. Linter (if configured) — fix findings your change introduced.
+3. Build/typecheck — exit code 0.
+4. FULL test suite — not just your new tests; regressions in untouched code are your problem if your change caused them.
+
+Capture commands and exit codes verbatim — they go in the handoff as evidence. Pre-existing failures: note explicitly, don't fix, don't hide. Apply the 3-strike rule: after 3 failed fix attempts, stop, revert to working state, escalate.
+
+## Phase 7 - Review
+
+Self-review the diff with `code-review` + `ai-slop-remover` loaded. For broad steps (5+ files or crossing architectural boundaries), request an independent pass (`code-reviewer` agent) instead. Fix blockers; re-run Phase 6 after any fix.
+
+## Phase 8 - Handoff
+
+Gate: every todo is either done or explicitly deferred with a reason. No silent drops.
+
+Write the handoff per `handoff-protocol` — schema, pasted evidence, deviations, downstream updates, notes for the next step. Append durable, step-independent facts to `plans/NOTES.md`. Set the plan's frontmatter `status: complete`.
+
+## Phase 9 - User approval
+
+Present: what was done, deviations, downstream plan changes (made or proposed), evidence summary, handoff location. Then STOP — do not begin the next step. If the user requests changes, address them, re-run Phase 6, update the handoff, and present again.
+
+## Anti-patterns
+
+- Editing code before the staleness check — the primary source of mid-step surprises
+- Implementing "while I'm here" improvements outside the plan's scope
+- Tests derived from the implementation instead of the acceptance criteria
+- Collecting build/test evidence BEFORE formatting/linting, then shipping different bytes
+- Running only your new tests and claiming "tests pass"
+- Silently absorbing a major deviation instead of escalating or proposing a plan diff
+- Rewriting downstream plan scope directly instead of proposing per `handoff-protocol`
+- Starting the next step without user approval

config.agent.example.yaml

@@ -7,12 +7,15 @@
 #   - <agent-name>_TOP_P
 #   - <agent-name>_GLOBAL_TOOLS (as a JSON string array)
 #   - <agent-name>_MCP_SERVERS (as a JSON string array)
+#   - <agent-name>_SPAWNABLE_AGENTS (as a JSON string array; see spawnable_agents below)
 #   - <agent-name>_AGENT_SESSION
 #   - <agent-name>_VARIABLES (as JSON array of key-value pairs; e.g. '[{"name": "username", "value": "alex"}]')
 
 model: openai:gpt-4o             # Specify the LLM to use
 temperature: null                # Set default temperature parameter, range (0, 1)
 top_p: null                      # Set default top-p parameter, with a range of (0, 1) or (0, 2) depending on the model
+reasoning_effort: null           # Reasoning effort level for models that support it (e.g. low, medium, high).
+                                 # Only valid when the agent's model declares reasoning_levels.
 agent_session: null              # Set a session to use when starting the agent. (e.g. temp, default); defaults to globally set agent_session
 name: <agent-name>               # Name of the agent, used in the UI and logs
 description: <description>       # Description of the agent, used in the UI
@@ -30,6 +33,12 @@ continuation_prompt: null        # Custom prompt used when auto-continuing (opti
 # Enable this agent to spawn and manage child agents in parallel.
 # See https://github.com/Dark-Alex-17/coyote/wiki/Agents for detailed documentation.
 can_spawn_agents: false          # Enable the agent to spawn child agents
+# spawnable_agents:              # Optional whitelist restricting which agents can be spawned via `agent__spawn`.
+#   - explore                    # If omitted (the default), ALL installed agents are spawnable. This is the unrestricted default.
+#   - coder                      # Provide a list to restrict. Match is exact and case-sensitive (use directory names).
+#   - oracle                     # An empty list ([]) means literally nothing spawnable.
+                                 # Also filters `agent__list_available` output so the LLM only sees what it can spawn.
+                                 # Graph agents (graph.yaml) ignore this; they declare spawn targets in agent nodes.
 max_concurrent_agents: 4         # Maximum number of agents that can run simultaneously
 max_agent_depth: 3               # Maximum nesting depth for sub-agents (prevents runaway spawning)
 inject_spawn_instructions: true  # Inject the default agent spawning instructions into the agent's system prompt
@@ -48,6 +57,12 @@ enabled_skills:                  # Optional list of skills available when this a
                                  # Must be a subset of global `visible_skills`. Omit to inherit the global default.
   - git-master
   - ai-slop-remover
+inject_skill_instructions: true  # Inject a short hint pointing the model at `skill__list` when skills are enabled
+                                 # (default: true). Suppressed automatically when no skills are available.
+skill_instructions: null         # Custom text for the skill hint (optional; uses built-in default if null)
+memory: null                     # Per-agent memory override (default: inherit). Set to `false` to disable memory
+                                 # for this agent regardless of workspace/global presence. See the Memory wiki page.
+
 dynamic_instructions: false      # Whether to use dynamic instructions for the agent; if false, static instructions are used
 instructions: |                  # Static instructions for the agent; ignored if dynamic instructions are used
   You are a AI agent designed to demonstrate agent capabilities.
@@ -86,6 +101,9 @@ conversation_starters:          # Optional conversation starters for the agent
   - What is the best way to exercise?
   - How do I manage my time effectively?
 documents:                     # Optional documents to load for the agent
+                               # To enable graph-based RAG (entity/relationship extraction + knowledge graph retrieval),
+                               # set `rag_extractor_model` in your global config.yaml.
+                               # See https://github.com/Dark-Alex-17/coyote/wiki/RAG#graph-based-rag
   - git:/some/repo             # Explicitly tell Coyote to use the 'git' document loader using an absolute path
   - pdf:some-pdf-file.pdf      # Explicitly tell Coyote to use the 'pdf' document loader using a relative path
   - https://some-website.com/some-page

config.example.yaml

@@ -2,6 +2,8 @@
 model: openai:gpt-4o             # Specify the LLM to use
 temperature: null                # Set default temperature parameter (0, 1)
 top_p: null                      # Set default top-p parameter, with a range of (0, 1) or (0, 2) depending on the model
+reasoning_effort: null           # Reasoning effort level for models that support it (e.g. low, medium, high).
+                                 # Only valid when the active model declares reasoning_levels. See the Clients docs.
 
 # ---- Behavior ----
 stream: true                     # Controls whether to use the stream-style APIs when querying for completions from LLM clients.
@@ -18,6 +20,7 @@ agent_session: null              # Set a session to use when starting an agent (
 
 # ---- Appearance ----
 highlight: true                  # Controls syntax highlighting
+raw_markdown: false              # When true, render markdown as raw text with syntax highlighting only. When false (default), transforms markdown syntax (headings, bold, lists, etc.) into styled terminal output
 light_theme: false               # Activates a light color theme when true. env: COYOTE_LIGHT_THEME
 
 # ---- Miscellaneous ----
@@ -31,7 +34,7 @@ sync_models_url: >               # URL to sync model changes from
 left_prompt:
   '{color.red}{model}){color.green}{?session {?agent {agent}>}{session}{?role /}}{!session {?agent {agent}>}}{role}{?rag @{rag}}{color.cyan}{?session )}{!session >}{color.reset} '
 right_prompt:
-  '{color.purple}{?session {?consume_tokens {consume_tokens}({consume_percent}%)}{!consume_tokens {consume_tokens}}}{color.reset}'
+  '{color.cyan}{?reasoning_effort [{reasoning_effort}] }{color.purple}{?session {?consume_tokens {consume_tokens}({consume_percent}%)}{!consume_tokens {consume_tokens}}}{color.reset}'
 
 # ---- Vault ----
 # See the [Vault documentation](https://github.com/Dark-Alex-17/coyote/wiki/Vault) for more information on the Coyote vault.
@@ -91,6 +94,7 @@ enabled_tools: null              # Which tools to enable by default.
                                  # Example (comma-separated form):
                                  #   enabled_tools: fs,web_search_coyote
 visible_tools:                   # Which tools are visible to be compiled (and are thus able to be defined in 'enabled_tools')
+#  - ast_grep.sh
 #  - demo_py.py
 #  - demo_sh.sh
 #  - demo_ts.ts
@@ -133,6 +137,14 @@ enabled_mcp_servers: null        # Which MCP servers to enable by default.
                                  #     - slack
                                  # Example (comma-separated form):
                                  #   enabled_mcp_servers: github,slack,ddg-search
+no_workspace_mcp: false          # Disable loading workspace-local MCP servers (default: false).
+                                 # When false (the default), Coyote merges the first workspace MCP config it finds
+                                 # into the global MCP registry at startup, checking in order:
+                                 #   1. .coyote/mcp.json
+                                 #   2. .coyote/.mcp.json  (Claude-style file name)
+                                 #   3. .mcp.json          (project root; Claude Code convention)
+                                 # Workspace entries shadow global ones on name collision.
+                                 # Set to true (or pass --no-workspace-mcp) to skip this entirely.
 
 # ---- Skills ----
 # Skills are modular knowledge or capability packs the LLM can load and unload mid-conversation.
@@ -152,6 +164,10 @@ enabled_skills: null             # Which skills are available by default (no rol
                                   #     - ai-slop-remover
                                   # Example (comma-separated form):
                                   #   enabled_skills: git-master,ai-slop-remover
+inject_skill_instructions: true   # Inject a short hint pointing the model at `skill__list` when skills are enabled in
+                                  # this context. Only injected if `function_calling_support`, `skills_enabled`, and the
+                                  # effective enabled skill set is non-empty (default: true).
+skill_instructions: null          # Custom text used for the skill hint when injected. If null, uses built-in default.
 
 # ---- Auto-Continue (Todo System) ----
 # The auto-continue system provides built-in task tracking for improved reliability.
@@ -171,6 +187,33 @@ summarization_prompt: >          # The text prompt used for creating a concise s
   'Summarize the discussion briefly in 200 words or less to use as a prompt for future context.'
 summary_context_prompt: >        # The text prompt used for including the summary of the entire session as context to the model
   'This is a summary of the chat history as a recap: '
+compression_keep_last: 0         # Number of most-recent messages to keep visible after compression (0 = compress all messages)
+max_tool_result_chars: null      # Cap on tool result characters forwarded to the model per call (null = no cap)
+
+# ---- Memory ----
+# See the [Memory documentation](https://github.com/Dark-Alex-17/coyote/wiki/Memory) for more information.
+# Memory is opt-in by workspace presence (`.coyote/memory/MEMORY.md`) and global
+# presence (`<config_dir>/memory/MEMORY.md`). Set `memory: false` to disable
+# even when memory files exist. The cascade is: agent > session > role > app.
+# Bootstrap with `coyote --init-memory [global|workspace]` to create the marker file
+# the LLM needs before it will write any memory.
+memory: null                     # null = enabled when memory exists on disk; true = force on; false = force off
+memory_cap_with_tools: null      # Char cap for injected memory when function calling is available (default: 6000).
+                                 # Only MEMORY.md indexes are injected; the LLM uses memory__read to fetch drill files.
+memory_cap_without_tools: null   # Char cap when function calling is unavailable (default: 12000).
+                                 # Indexes plus drill file bodies are injected up to this cap.
+
+# ---- Workspace Instructions ----
+# Human-curated project instructions injected read-only into the system prompt, in full.
+# Coyote walks up from the current directory and injects the first match from the file
+# chain below (per directory, in order). Scaffold with `coyote --init-instructions`.
+# Disable per-invocation with --no-workspace-instructions, or override the chain with
+# repeatable --workspace-instructions-file flags.
+workspace_instructions: null     # null/true = inject when an instructions file exists; false = never inject
+workspace_instructions_files: null # File name chain to search, in priority order.
+                                   # Default: [COYOTE.md, AGENTS.md, CLAUDE.md, GEMINI.md]
+                                   # Set to a custom list to reorder or drop fallbacks, e.g.:
+                                   #   workspace_instructions_files: [COYOTE.md]
 
 # ---- RAG ----
 # See the [RAG Docs](https://github.com/Dark-Alex-17/coyote/wiki/RAG) for more details.
@@ -179,6 +222,9 @@ rag_reranker_model: null         # Specifies the reranker model used for sorting
 rag_top_k: 5                     # Specifies the number of documents to retrieve for answering queries
 rag_chunk_size: null             # Defines the size of chunks for document processing in characters
 rag_chunk_overlap: null          # Defines the overlap between chunks
+rag_extractor_model: null        # LLM model for graph-based entity/relationship extraction; when set, enables a graph RAG signal alongside vector and BM25
+rag_extractor_prompt: null       # Custom extraction prompt template; must contain __CHUNK__ placeholder; defaults to built-in prompt when null
+rag_graph_hops: 1                # Number of hops to expand from matched entities at query time (0 = seed nodes only; 1 = direct neighbors; increase for denser graphs)
 # Defines the query structure using variables like __CONTEXT__, __SOURCES__, and __INPUT__ to tailor searches to specific needs
 rag_template: |
   Answer the query based on the context while respecting the rules. (user query, some textual context and rules, all inside xml tags)
@@ -305,11 +351,38 @@ clients:
     api_base: https://api.mistral.ai/v1
     api_key: '{{MISTRAL_API_KEY}}'                      # You can either hard-code or inject secrets from the Coyote vault
 
-  # See https://docs.x.ai/docs
+  # See https://docs.x.ai/docs - OAuth via SuperGrok / X Premium+ subscription
   - type: openai-compatible
     name: xai
     api_base: https://api.x.ai/v1
     api_key: '{{XAI_API_KEY}}'                          # You can either hard-code or inject secrets from the Coyote vault
+    auth: null                                          # When set to 'oauth', Coyote will use OAuth instead of an API key
+                                                        # Authenticate with `coyote --authenticate` or `.authenticate` in the REPL
+                                                        # Note: Oauth requires SuperGrok/X Premium+ subscription
+
+  # Example: private OpenAI-compatible gateway with client_credentials OAuth
+  # - type: openai-compatible
+  #   name: acme-gateway
+  #   api_base: https://gateway.acme.com/v1
+  #   auth: oauth
+  #   oauth:
+  #     client_id: '{{ACME_CLIENT_ID}}'
+  #     client_secret: '{{ACME_CLIENT_SECRET}}'
+  #     token_url: https://auth.acme.com/oauth/token
+  #     scopes: [openai.chat]
+  #     flow: client_credentials
+
+  # Example: OAuth via Device Authorization Grant (RFC 8628 — for CLIs like Moonshot's kimi-code, MiniMax mmx, etc.)
+  # - type: openai-compatible
+  #   name: moonshot
+  #   api_base: https://api.kimi.com/coding/v1
+  #   auth: oauth
+  #   oauth:
+  #     client_id: '{{MOONSHOT_CLIENT_ID}}'
+  #     device_authorization_url: https://auth.kimi.com/api/oauth/device_authorization
+  #     token_url: https://auth.kimi.com/api/oauth/token
+  #     flow: device_code
+  #     # use_pkce_in_device_flow: true    # enable if your provider requires PKCE with device flow
 
   # See https://docs.ai21.com/docs/overview
   - type: openai-compatible

config.role.example.md

@@ -8,6 +8,8 @@ name: <role-name>                     # The name of the role
 model: openai:gpt-4o                  # The model to use for this role
 temperature: 0.2                      # The temperature to use for this role when querying the model
 top_p: 0                              # The top_p to use for this role when querying the model
+reasoning_effort: null                # Reasoning effort level for models that support it (e.g. low, medium, high).
+                                      # Only valid when the role's model declares reasoning_levels.
 enabled_tools:                        # Tools to enable for this role. Accepts a YAML list (preferred)
   - fs_ls                             # or a comma-separated string (e.g. `enabled_tools: fs_ls,fs_cat`).
   - fs_cat                            # Use `all` to enable every visible tool.
@@ -19,6 +21,12 @@ skills_enabled: true                  # Master switch for skills in this role (d
 enabled_skills:                       # Skills available when this role is active. Accepts a YAML list (preferred)
   - git-master                        # or a comma-separated string (e.g. `enabled_skills: git-master,ai-slop-remover`).
   - ai-slop-remover                   # Must be a subset of global `visible_skills`. Omit to inherit the global default.
+inject_skill_instructions: true       # Inject a short hint pointing the model at `skill__list` when skills are enabled 
+                                      # (default: true). Suppressed automatically when no skills are available.
+skill_instructions: null              # Custom text for the skill hint (optional; uses built-in default if null)
+memory: null                          # Per-role memory override (default: inherit). Set to `false` to disable memory
+                                      # when this role is active. See the Memory wiki page.
+
 prompt: null                          # A custom prompt to use for this role that will immediately query
                                       # the model for output instead of using the instructions below
 # Auto-Continue (Todo System)

graph.example.yaml

@@ -33,6 +33,8 @@ version: "1.0"                     # Graph schema version. Only "1.0" is accepte
 model: claude:claude-sonnet-4-6    # Default model for `llm` nodes that don't override it
 temperature: 0.0                   # Default sampling temperature for `llm` nodes
 top_p: null                        # Default sampling top-p for `llm` nodes
+reasoning_effort: null             # Default reasoning effort for `llm` nodes that don't override it.
+                                   # Only valid when the model declares reasoning_levels.
 
 global_tools:                      # Tool universe an `llm` node's `tools:` whitelist draws from
   - web_search_coyote.sh
@@ -63,6 +65,9 @@ enabled_skills:
   - code-review
   - git-master
   - ai-slop-remover
+inject_skill_instructions: true   # Inject a hint pointing the model at `skill__list`. Defaults to true; suppressed
+                                  # automatically when no skills are available.
+skill_instructions: null          # Custom text for the skill hint (optional; uses the built-in default if omitted).
 
 conversation_starters:             # Suggested prompts surfaced in the UI
   - "Research the current state of WebAssembly outside the browser"
@@ -173,8 +178,12 @@ nodes:
     # catches violations at load time). `skills_enabled: false` would
     # disable skills entirely for this node (no meta-tools exposed).
     # Nothing is auto-loaded: the model decides when to load a skill.
+    skills_enabled: true       # Whether skills are enabled on this llm node; defaults to 'true'
     enabled_skills:
       - ai-slop-remover
+    inject_skill_instructions: true   # Override skill-hint injection for just this node. Falls back to
+                                      # agent/graph/global default when omitted.
+    skill_instructions: null          # Per-node skill-hint text override; uses the built-in default when omitted.
     output_schema:             # Optional JSON Schema. The output is parsed to JSON
       type: object             # and its top-level object keys auto-merge into state
       properties:              # (so `topic` / `needs_deep_dive` become {{topic}} etc).
@@ -218,6 +227,9 @@ nodes:
     chunk_size: 1000
     chunk_overlap: 100
     reranker_model: null       # Optional reranker for hybrid-search results
+    extractor_model: null      # Optional chat model for graph-based entity/relationship extraction; enables graph RAG signal when set
+    extractor_prompt: null     # Optional custom extraction prompt; must contain __CHUNK__ placeholder; uses built-in prompt when null
+    graph_hops: 1              # Graph expansion depth at query time (0 = seed nodes only; 1 = direct neighbors; increase for denser knowledge graphs)
     batch_size: 100            # Optional embedding-request batch size
     state_updates:             # {{output}} = { context: <str>, sources: [<path>, ...] }
       context: "{{output.context}}"   # writes `context` -> `reducers.context = concat`
@@ -384,6 +396,7 @@ nodes:
       - mcp:ddg-search         #   `mcp:<server>` includes that server's functions
     model: claude:claude-haiku-4-5  # Optional per-node model override
     temperature: 0.3           # Optional per-node sampling override
+    reasoning_effort: null     # Optional per-node reasoning effort override (e.g. low, medium, high)
     max_attempts: 2            # Retry count on transient errors only. Default 1.
     max_iterations: 10         # Tool-call-loop turn cap. Default 10.
     fallback: review           # Route here if all attempts fail

justfile

@@ -23,3 +23,16 @@ fmt:
 [arg('build_type', pattern="debug|release")]
 build build_type='debug':
 	@cargo build {{ if build_type == "release" { "--release" } else { "" } }}
+
+# Build a multi-platform Docker image (linux/amd64 + linux/arm64).
+# Requires an active buildx builder with multi-platform support and a registry login.
+# version: must match an existing GitHub release tag (e.g. 0.7.4)
+# image:   registry/image name to push to (default: darkalex17/coyote)
+[group: 'build']
+docker-build version image='darkalex17/coyote':
+	docker buildx build \
+		--platform linux/amd64,linux/arm64 \
+		--build-arg COYOTE_VERSION={{ version }} \
+		--tag {{ image }}:{{ version }} \
+		--tag {{ image }}:latest \
+		.

models.yaml

@@ -3,6 +3,33 @@
 #  - https://platform.openai.com/docs/api-reference/chat
 - provider: openai
   models:
+    - name: gpt-5.6-sol
+      max_input_tokens: 1050000
+      max_output_tokens: 128000
+      input_price: 5
+      output_price: 30
+      supports_vision: true
+      supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh, max]
+      default_reasoning_effort: medium
+    - name: gpt-5.6-terra
+      max_input_tokens: 1050000
+      max_output_tokens: 128000
+      input_price: 5
+      output_price: 30
+      supports_vision: true
+      supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh, max]
+      default_reasoning_effort: medium
+    - name: gpt-5.6-luna
+      max_input_tokens: 1050000
+      max_output_tokens: 128000
+      input_price: 5
+      output_price: 30
+      supports_vision: true
+      supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh, max]
+      default_reasoning_effort: medium
     - name: gpt-5.5
       max_input_tokens: 1050000
       max_output_tokens: 128000
@@ -10,6 +37,8 @@
       output_price: 30
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh]
+      default_reasoning_effort: medium
     - name: gpt-5.5-pro
       max_input_tokens: 1050000
       max_output_tokens: 128000
@@ -17,6 +46,8 @@
       output_price: 180
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [medium, high, xhigh]
+      default_reasoning_effort: high
     - name: gpt-5.4
       max_input_tokens: 1050000
       max_output_tokens: 128000
@@ -24,6 +55,8 @@
       output_price: 15
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh]
+      default_reasoning_effort: none
     - name: gpt-5.4-pro
       max_input_tokens: 1050000
       max_output_tokens: 128000
@@ -31,6 +64,8 @@
       output_price: 180
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [medium, high, xhigh]
+      default_reasoning_effort: medium
     - name: gpt-5.4-mini
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -38,6 +73,8 @@
       output_price: 4.5
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh]
+      default_reasoning_effort: none
     - name: gpt-5.4-nano
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -45,6 +82,8 @@
       output_price: 1.25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh]
+      default_reasoning_effort: none
     - name: gpt-5.3-codex
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -52,6 +91,8 @@
       output_price: 14
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh]
+      default_reasoning_effort: medium
     - name: chat-latest
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -66,6 +107,17 @@
       output_price: 14
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh]
+      default_reasoning_effort: none
+    - name: gpt-5.2-pro
+      max_input_tokens: 400000
+      max_output_tokens: 128000
+      input_price: 21
+      output_price: 168
+      supports_vision: true
+      supports_function_calling: true
+      reasoning_levels: [medium, high, xhigh]
+      default_reasoning_effort: medium
     - name: gpt-5.1
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -73,6 +125,8 @@
       output_price: 10
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high]
+      default_reasoning_effort: none
     - name: gpt-5.1-chat-latest
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -80,6 +134,8 @@
       output_price: 10
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high]
+      default_reasoning_effort: none
     - name: gpt-5
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -87,6 +143,8 @@
       output_price: 10
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: medium
     - name: gpt-5-chat-latest
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -94,6 +152,8 @@
       output_price: 10
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: medium
     - name: gpt-5-mini
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -151,6 +211,8 @@
       supports_vision: true
       supports_function_calling: true
       system_prompt_prefix: Formatting re-enabled
+      reasoning_levels: [low, medium, high]
+      default_reasoning_effort: medium
       patch:
         body:
           max_tokens: null
@@ -258,24 +320,38 @@
 #  - https://ai.google.dev/api/rest/v1beta/models/streamGenerateContent
 - provider: gemini
   models:
+    - name: gemini-3.6-flash
+      max_input_tokens: 1048576
+      max_output_tokens: 65536
+      input_price: 1.5
+      output_price: 7.5
+      supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_level: medium
     - name: gemini-3.5-flash
       max_input_tokens: 1048576
       max_output_tokens: 65536
       input_price: 0.2
       output_price: 1.5
       supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: medium
     - name: gemini-3-flash-preview
       max_input_tokens: 1048576
       max_output_tokens: 65536
       input_price: 0.2
       output_price: 1.5
       supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: high
     - name: gemini-3.1-flash-lite
       max_input_tokens: 1048576
       max_output_tokens: 65536
       input_price: 0.2
       output_price: 1.5
       supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: minimal
     - name: gemini-3.1-pro-preview
       max_input_tokens: 1048576
       max_output_tokens: 65535
@@ -283,6 +359,8 @@
       output_price: 2.5
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high]
+      default_reasoning_effort: high
     - name: gemini-2.5-flash
       max_input_tokens: 1048576
       max_output_tokens: 65536
@@ -297,6 +375,8 @@
       output_price: 0
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high]
+      default_reasoning_effort: high
     - name: gemini-2.5-flash-lite
       max_input_tokens: 1000000
       max_output_tokens: 64000
@@ -308,10 +388,14 @@
       max_input_tokens: 1048576
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, high]
+      default_reasoning_level: high
     - name: gemini-3-flash-preview
       max_input_tokens: 1048576
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_level: high
     - name: gemma-3-27b-it
       max_input_tokens: 131072
       max_output_tokens: 8192
@@ -329,6 +413,16 @@
 #  - https://docs.anthropic.com/en/api/messages
 - provider: claude
   models:
+    - name: claude-fable-5
+      max_input_tokens: 1000000
+      max_output_tokens: 128000
+      require_max_tokens: true
+      input_price: 10
+      output_price: 50
+      supports_function_calling: true
+      supports_vision: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: claude-opus-4-8
       max_input_tokens: 1000000
       max_output_tokens: 128000
@@ -337,6 +431,8 @@
       output_price: 25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: claude-opus-4-7
       max_input_tokens: 1000000
       max_output_tokens: 128000
@@ -345,6 +441,8 @@
       output_price: 25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: claude-opus-4-6
       max_input_tokens: 200000
       max_output_tokens: 8192
@@ -353,6 +451,8 @@
       output_price: 25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, max]
+      default_reasoning_effort: high
     - name: claude-opus-4-6:thinking
       real_name: claude-opus-4-6
       max_input_tokens: 200000
@@ -369,6 +469,16 @@
           thinking:
             type: enabled
             budget_tokens: 16000
+    - name: claude-sonnet-5
+      max_input_tokens: 1000000
+      max_output_tokens: 128000
+      require_max_tokens: true
+      input_price: 3
+      output_price: 15
+      supports_vision: true
+      supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: claude-sonnet-4-6
       max_input_tokens: 200000
       max_output_tokens: 8192
@@ -377,6 +487,8 @@
       output_price: 15
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, max]
+      default_reasoning_effort: high
     - name: claude-sonnet-4-6:thinking
       real_name: claude-sonnet-4-6
       max_input_tokens: 200000
@@ -700,14 +812,65 @@
 #  - https://docs.x.ai/docs/models
 #  - https://docs.x.ai/docs/api-reference#chat-completions
 - provider: xai
+  oauth:
+    client_id: b1a00492-073a-47ea-816f-4c329264a828
+    authorize_url: https://auth.x.ai/oauth2/authorize
+    token_url: https://auth.x.ai/oauth2/token
+    scopes:
+      - openid
+      - profile
+      - email
+      - offline_access
+      - grok-cli:access
+      - api:access
+    redirect_port: 56121
+    flow: pkce
+    token_request_format: form_url_encoded
+    extra_authorize_params:
+      plan: generic
+      referrer: coyote
+    echo_pkce_in_token_exchange: true
   models:
+    - name: grok-4.5
+      input_price: 2
+      output_price: 6
+      max_input_tokens: 256000
+      supports_function_calling: true
+    - name: grok-build-0.1
+      input_price: 1
+      output_price: 2
+      max_input_tokens: 256000
+      supports_function_calling: true
+    - name: grok-4.3
+      input_price: 1.25
+      output_price: 2.5
+      max_input_tokens: 1000000
+      supports_function_calling: true
+    - name: grok-4.20
+      real_name: grok-4.20-multi-agent-0309
+      input_price: 1.25
+      output_price: 2.5
+      max_input_tokens: 1000000
+      supports_function_calling: true
+    - name: grok-4.20-reasoning
+      real_name: grok-4.20-0309-reasoning
+      input_price: 1.25
+      output_price: 2.5
+      max_input_tokens: 1000000
+      supports_function_calling: true
+    - name: grok-4.20-non-reasoning
+      real_name: grok-4.20-0309-non-reasoning
+      input_price: 1.25
+      output_price: 2.5
+      max_input_tokens: 1000000
+      supports_function_calling: true
     - name: grok-4-1-fast-non-reasoning
-      max_input_tokens: 2000000
+      max_input_tokens: 1000000
       input_price: 0.2
       output_price: 0.5
       supports_function_calling: true
     - name: grok-4-1-fast-reasoning
-      max_input_tokens: 2000000
+      max_input_tokens: 1000000
       input_price: 0.2
       output_price: 0.5
       supports_function_calling: true
@@ -819,18 +982,24 @@
       input_price: 0.2
       output_price: 1.5
       supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: medium
     - name: gemini-3-flash-preview
       max_input_tokens: 1048576
       max_output_tokens: 65536
       input_price: 0.2
       output_price: 1.5
       supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: high
     - name: gemini-3.1-flash-lite
       max_input_tokens: 1048576
       max_output_tokens: 65536
       input_price: 0.2
       output_price: 1.5
       supports_function_calling: true
+      reasoning_levels: [minimal, high]
+      default_reasoning_effort: minimal
     - name: gemini-3.1-pro-preview
       max_input_tokens: 1048576
       max_output_tokens: 65536
@@ -838,6 +1007,8 @@
       output_price: 12
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high]
+      default_reasoning_effort: high
     - name: gemini-2.5-flash
       max_input_tokens: 1048576
       max_output_tokens: 65535
@@ -845,6 +1016,8 @@
       output_price: 2.5
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high]
+      default_reasoning_effort: medium
     - name: gemini-2.5-pro
       max_input_tokens: 1048576
       max_output_tokens: 65536
@@ -852,6 +1025,8 @@
       output_price: 10
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high]
+      default_reasoning_effort: high
     - name: gemini-2.5-flash-lite
       max_input_tokens: 1048576
       max_output_tokens: 65536
@@ -863,10 +1038,24 @@
       max_input_tokens: 1048576
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, high]
+      default_reasoning_effort: high
     - name: gemini-3-flash-preview
       max_input_tokens: 1048576
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: high
+    - name: claude-fable-5
+      max_input_tokens: 1000000
+      max_output_tokens: 128000
+      require_max_tokens: true
+      input_price: 10
+      output_price: 50
+      supports_function_calling: true
+      supports_vision: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: claude-opus-4-8
       max_input_tokens: 1000000
       max_output_tokens: 128000
@@ -875,6 +1064,8 @@
       output_price: 25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: claude-opus-4-7
       max_input_tokens: 1000000
       max_output_tokens: 128000
@@ -883,6 +1074,8 @@
       output_price: 25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: claude-opus-4-6
       max_input_tokens: 200000
       max_output_tokens: 8192
@@ -906,6 +1099,16 @@
           thinking:
             type: enabled
             budget_tokens: 16000
+    - name: claude-sonnet-5
+      max_input_tokens: 1000000
+      max_output_tokens: 128000
+      require_max_tokens: true
+      input_price: 3
+      output_price: 15
+      supports_vision: true
+      supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: claude-sonnet-4-6
       max_input_tokens: 200000
       max_output_tokens: 8192
@@ -914,6 +1117,8 @@
       output_price: 15
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, max]
+      default_reasoning_effort: high
     - name: claude-sonnet-4-6:thinking
       real_name: claude-sonnet-4-6
       max_input_tokens: 200000
@@ -1038,6 +1243,16 @@
 #  - https://docs.aws.amazon.com/bedrock/latest/userguide/conversation-inference-call.html
 - provider: bedrock
   models:
+    - name: us.anthropic.claude-fable-5
+      max_input_tokens: 1000000
+      max_output_tokens: 128000
+      require_max_tokens: true
+      input_price: 10
+      output_price: 50
+      supports_function_calling: true
+      supports_vision: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: us.anthropic.claude-opus-4-8
       max_input_tokens: 1000000
       max_output_tokens: 128000
@@ -1046,6 +1261,8 @@
       output_price: 25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: us.anthropic.claude-opus-4-7
       max_input_tokens: 1000000
       max_output_tokens: 128000
@@ -1054,6 +1271,8 @@
       output_price: 25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: us.anthropic.claude-opus-4-6-v1
       max_input_tokens: 200000
       max_output_tokens: 8192
@@ -1062,6 +1281,8 @@
       output_price: 25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, max]
+      default_reasoning_effort: high
     - name: us.anthropic.claude-opus-4-6-v1:thinking
       real_name: us.anthropic.claude-opus-4-6-v1
       max_input_tokens: 200000
@@ -1079,6 +1300,16 @@
             thinking:
               type: enabled
               budget_tokens: 16000
+    - name: us.anthropic.claude-sonnet-5
+      max_input_tokens: 1000000
+      max_output_tokens: 128000
+      require_max_tokens: true
+      input_price: 3
+      output_price: 15
+      supports_vision: true
+      supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: us.anthropic.claude-sonnet-4-6
       max_input_tokens: 200000
       max_output_tokens: 8192
@@ -1087,6 +1318,8 @@
       output_price: 15
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, max]
+      default_reasoning_effort: high
     - name: us.anthropic.claude-sonnet-4-6:thinking
       real_name: us.anthropic.claude-sonnet-4-6
       max_input_tokens: 200000
@@ -1468,6 +1701,30 @@
 #  - https://platform.moonshot.cn/docs/api/chat#%E5%85%AC%E5%BC%80%E7%9A%84%E6%9C%8D%E5%8A%A1%E5%9C%B0%E5%9D%80
 - provider: moonshot
   models:
+    - name: kimi-k3
+      max_input_tokens: 1048576
+      input_price: 3
+      output_price: 15
+      supports_vision: true
+      supports_function_calling: true
+    - name: kimi-k2.7-code
+      max_input_tokens: 262144
+      input_price: 0.95
+      output_price: 4
+      supports_vision: true
+      supports_function_calling: true
+    - name: kimi-k2.7-code-highspeed
+      max_input_tokens: 262144
+      input_price: 1.9
+      output_price: 8
+      supports_vision: true
+      supports_function_calling: true
+    - name: kimi-k2.6
+      max_input_tokens: 262144
+      input_price: 0.95
+      output_price: 4
+      supports_vision: true
+      supports_function_calling: true
     - name: kimi-k2.5
       max_input_tokens: 262144
       input_price: 0.56
@@ -1502,6 +1759,18 @@
 #  - https://platform.deepseek.com/api-docs/api/create-chat-completion
 - provider: deepseek
   models:
+    - name: deepseek-v4-pro
+      max_input_tokens: 1000000
+      max_output_tokens: 384000
+      input_price: 0.435
+      output_price: 0.87
+      supports_function_calling: true
+    - name: deepseek-v4-flash
+      max_input_tokens: 1000000
+      max_output_tokens: 384000
+      input_price: 0.14
+      output_price: 0.28
+      supports_function_calling: true
     - name: deepseek-chat
       max_input_tokens: 64000
       max_output_tokens: 8192
@@ -1570,6 +1839,16 @@
 # - https://platform.minimaxi.com/document/ChatCompletion%20v2
 - provider: minimax
   models:
+    - name: minimax-m3
+      max_input_tokens: 1000000
+      input_price: 4.2
+      output_price: 16.8
+      supports_function_calling: true
+    - name: minimax-m2.7
+      max_input_tokens: 204800
+      input_price: 0.294
+      output_price: 1.176
+      supports_function_calling: true
     - name: minimax-m2.5
       max_input_tokens: 204800
       input_price: 0.294
@@ -1596,6 +1875,33 @@
 #  - https://openrouter.ai/docs/api-reference/chat-completion
 - provider: openrouter
   models:
+    - name: openai/gpt-5.6-sol
+      max_input_tokens: 1050000
+      max_output_tokens: 128000
+      input_price: 5
+      output_price: 30
+      supports_vision: true
+      supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh, max]
+      default_reasoning_effort: medium
+    - name: openai/gpt-5.6-terra
+      max_input_tokens: 1050000
+      max_output_tokens: 128000
+      input_price: 5
+      output_price: 30
+      supports_vision: true
+      supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh, max]
+      default_reasoning_effort: medium
+    - name: openai/gpt-5.6-luna
+      max_input_tokens: 1050000
+      max_output_tokens: 128000
+      input_price: 5
+      output_price: 30
+      supports_vision: true
+      supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh, max]
+      default_reasoning_effort: medium
     - name: openai/gpt-5.5
       max_input_tokens: 1050000
       max_output_tokens: 128000
@@ -1603,6 +1909,8 @@
       output_price: 30
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh]
+      default_reasoning_effort: medium
     - name: openai/gpt-5.5-pro
       max_input_tokens: 1050000
       max_output_tokens: 128000
@@ -1610,6 +1918,8 @@
       output_price: 180
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [medium, high, xhigh]
+      default_reasoning_effort: high
     - name: openai/gpt-5.4
       max_input_tokens: 1050000
       max_output_tokens: 128000
@@ -1617,6 +1927,8 @@
       output_price: 15
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh]
+      default_reasoning_effort: none
     - name: openai/gpt-5.4-pro
       max_input_tokens: 1050000
       max_output_tokens: 128000
@@ -1624,6 +1936,8 @@
       output_price: 180
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [medium, high, xhigh]
+      default_reasoning_effort: medium
     - name: openai/gpt-5.4-mini
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -1631,6 +1945,8 @@
       output_price: 4.5
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh]
+      default_reasoning_effort: none
     - name: openai/gpt-5.4-nano
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -1638,6 +1954,8 @@
       output_price: 1.25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh]
+      default_reasoning_effort: none
     - name: openai/gpt-5.3-codex
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -1645,6 +1963,8 @@
       output_price: 14
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh]
+      default_reasoning_effort: medium
     - name: openai/gpt-5.2
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -1652,6 +1972,17 @@
       output_price: 14
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh]
+      default_reasoning_effort: none
+    - name: openai/gpt-5.2-pro
+      max_input_tokens: 400000
+      max_output_tokens: 128000
+      input_price: 21
+      output_price: 168
+      supports_vision: true
+      supports_function_calling: true
+      reasoning_levels: [medium, high, xhigh]
+      default_reasoning_effort: medium
     - name: openai/gpt-5
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -1659,6 +1990,8 @@
       output_price: 10
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: medium
     - name: openai/gpt-5-mini
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -1696,18 +2029,67 @@
       input_price: 0.04
       output_price: 0.16
       supports_function_calling: true
+    - name: google/gemini-3.5-flash
+      max_input_tokens: 1048576
+      max_output_tokens: 65536
+      input_price: 0.2
+      output_price: 1.5
+      supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: medium
+    - name: google/gemini-3-flash-preview
+      max_input_tokens: 1048576
+      max_output_tokens: 65536
+      input_price: 0.2
+      output_price: 1.5
+      supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: high
+    - name: google/gemini-3.1-flash-lite
+      max_input_tokens: 1048576
+      max_output_tokens: 65536
+      input_price: 0.2
+      output_price: 1.5
+      supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: minimal
+    - name: google/gemini-3.1-pro-preview
+      max_input_tokens: 1048576
+      max_output_tokens: 65535
+      input_price: 0.3
+      output_price: 2.5
+      supports_vision: true
+      supports_function_calling: true
+      reasoning_levels: [low, medium, high]
+      default_reasoning_effort: high
+    - name: google/gemini-3-pro-preview
+      max_input_tokens: 1048576
+      supports_vision: true
+      supports_function_calling: true
+      reasoning_levels: [low, high]
+      default_reasoning_level: high
+    - name: google/gemini-3-flash-preview
+      max_input_tokens: 1048576
+      supports_vision: true
+      supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_level: high
     - name: google/gemini-2.5-flash
       max_input_tokens: 1048576
       input_price: 0.3
       output_price: 2.5
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high]
+      default_reasoning_effort: low
     - name: google/gemini-2.5-pro
       max_input_tokens: 1048576
       input_price: 1.25
       output_price: 10
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high]
+      default_reasoning_effort: high
     - name: google/gemini-2.5-flash-lite
       max_input_tokens: 1048576
       input_price: 0.3
@@ -1729,6 +2111,16 @@
       max_input_tokens: 131072
       input_price: 0.1
       output_price: 0.2
+    - name: anthropic/claude-fable-5
+      max_input_tokens: 1000000
+      max_output_tokens: 128000
+      require_max_tokens: true
+      input_price: 10
+      output_price: 50
+      supports_function_calling: true
+      supports_vision: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: anthropic/claude-opus-4-8
       max_input_tokens: 1000000
       max_output_tokens: 128000
@@ -1737,6 +2129,8 @@
       output_price: 25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: anthropic/claude-opus-4-7
       max_input_tokens: 1000000
       max_output_tokens: 128000
@@ -1745,6 +2139,8 @@
       output_price: 25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: anthropic/claude-opus-4.6
       max_input_tokens: 200000
       max_output_tokens: 8192
@@ -1753,6 +2149,18 @@
       output_price: 25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, max]
+      default_reasoning_effort: high
+    - name: anthropic/claude-sonnet-5
+      max_input_tokens: 1000000
+      max_output_tokens: 128000
+      require_max_tokens: true
+      input_price: 3
+      output_price: 15
+      supports_vision: true
+      supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: anthropic/claude-sonnet-4.6
       max_input_tokens: 200000
       max_output_tokens: 8192
@@ -1761,6 +2169,8 @@
       output_price: 15
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, max]
+      default_reasoning_effort: high
     - name: anthropic/claude-opus-4.5
       max_input_tokens: 200000
       max_output_tokens: 8192

scripts/install_coyote.ps1

@@ -39,7 +39,7 @@ switch ([System.Runtime.InteropServices.RuntimeInformation]::OSArchitecture) {
 
 if (-not $BinDir) {
   if ($isWin) { $BinDir = Join-Path $env:LOCALAPPDATA 'coyote\bin' }
-  else { $home = $env:HOME; if (-not $home) { $home = (Get-Item -Path ~).FullName }; $BinDir = Join-Path $home '.local/bin' }
+  else { $userHome = $env:HOME; if (-not $userHome) { $userHome = (Get-Item -Path ~).FullName }; $BinDir = Join-Path $userHome '.local/bin' }
 }
 New-Item -ItemType Directory -Force -Path $BinDir | Out-Null
 
@@ -95,13 +95,13 @@ if ($asset.name -match '\.zip$') {
   [System.IO.Compression.ZipFile]::ExtractToDirectory($archive, $extractDir)
 } elseif ($asset.name -match '\.tar\.gz$' -or $asset.name -match '\.tgz$') {
   $tar = Get-Command tar -ErrorAction SilentlyContinue
-  if ($tar) { & $tar.FullName -xzf $archive -C $extractDir }
+  if ($tar) { & $tar.Source -xzf $archive -C $extractDir }
   else { Fail "Asset is tar archive but 'tar' is not available." }
 } else {
   try { Add-Type -AssemblyName System.IO.Compression.FileSystem; [System.IO.Compression.ZipFile]::ExtractToDirectory($archive, $extractDir) }
   catch {
     $tar = Get-Command tar -ErrorAction SilentlyContinue
-    if ($tar) { & $tar.FullName -xf $archive -C $extractDir } else { Fail "Unknown archive format; neither zip nor tar workable." }
+    if ($tar) { & $tar.Source -xf $archive -C $extractDir } else { Fail "Unknown archive format; neither zip nor tar workable." }
   }
 }
 

scripts/install_coyote.sh

@@ -133,30 +133,21 @@ else
   echo "Error: unsupported OS for this installer: $OS" >&2; exit 1
 fi
 
+DL_URLS=$(grep -oE '"browser_download_url":[[:space:]]*"[^"]+"' "$JSON" \
+  | sed -E 's/.*"browser_download_url":[[:space:]]*"//; s/"$//' \
+  || true)
+
 ASSET_NAME=""; ASSET_URL=""
 for candidate in "${ASSET_CANDIDATES[@]}"; do
-  NAME=$(grep -oE '"name":\s*"[^"]+"' "$JSON" | sed 's/"name":\s*"//; s/"$//' | grep -Fx "$candidate" || true)
+  while IFS= read -r url; do
-  if [[ -n "$NAME" ]]; then
+    [[ -z "$url" ]] && continue
-    ASSET_NAME="$NAME"
+    if [[ "$url" == */"$candidate" ]]; then
-    ASSET_URL=$(awk -v pat="$NAME" '
+      ASSET_NAME="$candidate"
-      BEGIN{ FS=":"; want=0 }
+      ASSET_URL="$url"
-      /"name"/ {
+      break
-        line=$0;
-        gsub(/^\s+|\s+$/,"",line);
-        gsub(/"name"\s*:\s*"|"/ ,"", line);
-        want = (line==pat) ? 1 : 0;
-        next
-      }
-      want==1 && /"browser_download_url"/ {
-        u=$0;
-        gsub(/^\s+|\s+$/,"",u);
-        gsub(/.*"browser_download_url"\s*:\s*"|".*/ ,"", u);
-        print u;
-        exit
-      }
-    ' "$JSON")
-    if [[ -n "$ASSET_URL" ]]; then break; fi
     fi
+  done <<< "$DL_URLS"
+  [[ -n "$ASSET_URL" ]] && break
 done
 
 if [[ -z "$ASSET_URL" ]]; then

src/cli/completer.rs

@@ -5,9 +5,9 @@ use crate::utils::list_file_names;
 use crate::vault::Vault;
 use clap_complete::{CompletionCandidate, Shell, generate};
 use clap_complete_nushell::Nushell;
-use std::env;
 use std::ffi::OsStr;
 use std::io;
+use std::{env, fs};
 
 const COYOTE_CLI_NAME: &str = "coyote";
 
@@ -134,6 +134,34 @@ pub(super) fn session_completer(current: &OsStr) -> Vec<CompletionCandidate> {
         .collect()
 }
 
+pub(super) fn mcp_server_completer(current: &OsStr) -> Vec<CompletionCandidate> {
+    let cur = current.to_string_lossy();
+    let content = match fs::read_to_string(paths::mcp_config_file()) {
+        Ok(c) => c,
+        Err(_) => return vec![],
+    };
+    let json: serde_json::Value = match serde_json::from_str(&content) {
+        Ok(v) => v,
+        Err(_) => return vec![],
+    };
+    let servers = match json.get("mcpServers").and_then(|v| v.as_object()) {
+        Some(s) => s,
+        None => return vec![],
+    };
+
+    servers
+        .iter()
+        .filter(|(_, v)| {
+            v.get("type")
+                .and_then(|t| t.as_str())
+                .map(|t| t == "http" || t == "sse")
+                .unwrap_or(false)
+        })
+        .filter(|(k, _)| k.starts_with(&*cur))
+        .map(|(k, _)| CompletionCandidate::new(k))
+        .collect()
+}
+
 pub(super) fn secrets_completer(current: &OsStr) -> Vec<CompletionCandidate> {
     let cur = current.to_string_lossy();
     match load_app_config_for_completion() {

src/cli/mod.rs

@@ -1,12 +1,12 @@
 mod completer;
 
 use crate::cli::completer::{
-    ShellCompletion, agent_completer, macro_completer, model_completer, rag_completer,
+    ShellCompletion, agent_completer, macro_completer, mcp_server_completer, model_completer,
-    role_completer, secrets_completer, session_completer,
+    rag_completer, role_completer, secrets_completer, session_completer,
 };
-use crate::config::{AssetCategory, InstallFilter};
+use crate::config::{AssetCategory, InstallFilter, MemoryScope};
 use anyhow::{Context, Result};
-use clap::ValueHint;
+use clap::{ArgGroup, ValueHint};
 use clap::{Parser, crate_authors, crate_description, crate_version};
 use clap_complete::ArgValueCompleter;
 use is_terminal::IsTerminal;
@@ -27,9 +27,26 @@ use std::io::{Read, stdin};
 {usage-heading} {usage}
 
 {all-args}{after-help}
-"
+",
+	group(
+		ArgGroup::new("sbx-mode")
+			.args(["sandbox", "fresh", "no_mixins"])
+			.multiple(true)
+			.conflicts_with_all([
+				"model", "prompt", "role", "session", "agent", "rag", "rebuild_rag",
+				"macro_name", "execute", "code", "file", "no_stream", "no_memory",
+				"init_memory", "dry_run", "info", "build_tools", "install",
+				"install_from", "sync_models", "list_models", "list_roles",
+				"list_sessions", "list_agents", "list_rags", "list_macros",
+				"list_skills", "skill", "tail_logs", "completions", "update",
+			])
+	),
 )]
 pub struct Cli {
+    /// Input text
+    #[arg(trailing_var_arg = true)]
+    text: Vec<String>,
+
     /// Select a LLM model
     #[arg(short, long, add = ArgValueCompleter::new(model_completer))]
     pub model: Option<String>,
@@ -39,30 +56,6 @@ pub struct Cli {
     /// Select a role
     #[arg(short, long, add = ArgValueCompleter::new(role_completer))]
     pub role: Option<String>,
-    /// Start or join a session
-    #[arg(short = 's', long, add = ArgValueCompleter::new(session_completer))]
-    pub session: Option<Option<String>>,
-    /// Ensure the session is empty
-    #[arg(long)]
-    pub empty_session: bool,
-    /// Ensure the new conversation is saved to the session
-    #[arg(long)]
-    pub save_session: bool,
-    /// Start an agent
-    #[arg(short = 'a', long, add = ArgValueCompleter::new(agent_completer))]
-    pub agent: Option<String>,
-    /// Set agent variables
-    #[arg(long, value_names = ["NAME", "VALUE"], num_args = 2)]
-    pub agent_variable: Vec<String>,
-    /// Start a RAG
-    #[arg(long, add = ArgValueCompleter::new(rag_completer))]
-    pub rag: Option<String>,
-    /// Rebuild the RAG to sync document changes
-    #[arg(long)]
-    pub rebuild_rag: bool,
-    /// Execute a macro
-    #[arg(long = "macro", value_name = "MACRO", add = ArgValueCompleter::new(macro_completer))]
-    pub macro_name: Option<String>,
     /// Execute commands in natural language
     #[arg(short = 'e', long)]
     pub execute: bool,
@@ -75,92 +68,188 @@ pub struct Cli {
     /// Turn off stream mode
     #[arg(short = 'S', long)]
     pub no_stream: bool,
+    /// Render markdown as raw text with syntax highlighting only (skip the rich markdown renderer)
+    #[arg(long)]
+    pub raw_markdown: bool,
     /// Display the message without sending it
     #[arg(long)]
     pub dry_run: bool,
-    /// Display information
+    /// Disable loading workspace MCP servers from .coyote/mcp.json, .coyote/.mcp.json, or .mcp.json
     #[arg(long)]
-    pub info: bool,
+    pub no_workspace_mcp: bool,
-    /// Build all configured Bash tool scripts
+    /// Disable memory for this invocation
     #[arg(long)]
-    pub build_tools: bool,
+    pub no_memory: bool,
-    /// Reinstall bundled assets, overwriting any local changes
+    /// Disable loading workspace instructions (COYOTE.md/AGENTS.md/CLAUDE.md/etc.) for this invocation
-    #[arg(long, value_name = "CATEGORY", value_enum)]
-    pub install: Option<AssetCategory>,
-    /// Install assets from a remote git repository (URL may be suffixed with #<ref>)
-    #[arg(long, value_name = "GIT_URL")]
-    pub install_from: Option<String>,
-    /// Restrict --install-from to a single asset category
-    #[arg(long, value_name = "CATEGORY", value_enum, requires = "install_from")]
-    pub filter: Option<InstallFilter>,
-    /// Overwrite all conflicts without prompting (used with --install-from)
-    #[arg(long, requires = "install_from")]
-    pub install_force: bool,
-    /// Sync models updates
     #[arg(long)]
-    pub sync_models: bool,
+    pub no_workspace_instructions: bool,
-    /// List all available chat models
+    /// Override the workspace instructions file chain for this invocation (repeatable, priority order)
+    #[arg(long, value_name = "NAME")]
+    pub workspace_instructions_file: Vec<String>,
+    /// Skip permission prompts by setting AUTO_CONFIRM for all tools (dangerous!)
     #[arg(long)]
-    pub list_models: bool,
+    pub dangerously_skip_permissions: bool,
-    /// List all roles
+
-    #[arg(long)]
+    /// Start or join a session
-    pub list_roles: bool,
+    #[arg(short = 's', long, help_heading = "Session & Memory", add = ArgValueCompleter::new(session_completer))]
-    /// List all sessions
+    pub session: Option<Option<String>>,
-    #[arg(long)]
+    /// Ensure the session is empty
-    pub list_sessions: bool,
+    #[arg(long, help_heading = "Session & Memory")]
-    /// List all agents
+    pub empty_session: bool,
-    #[arg(long)]
+    /// Ensure the new conversation is saved to the session
-    pub list_agents: bool,
+    #[arg(long, help_heading = "Session & Memory")]
-    /// List all RAGs
+    pub save_session: bool,
-    #[arg(long)]
+    /// Bootstrap a memory marker so coyote begins loading memory next run
-    pub list_rags: bool,
+    #[arg(
-    /// List all macros
+        long,
-    #[arg(long)]
+        value_name = "SCOPE",
-    pub list_macros: bool,
+        value_enum,
-    /// List all installed skills
+        help_heading = "Session & Memory"
-    #[arg(long)]
+    )]
-    pub list_skills: bool,
+    pub init_memory: Option<MemoryScope>,
+    /// Scaffold a COYOTE.md workspace instructions file in the current directory
+    #[arg(long, help_heading = "Session & Memory")]
+    pub init_instructions: bool,
     /// Pre-load an existing skill into the session (repeatable). If a single
     /// `--skill <NAME>` is given and the skill doesn't exist, opens $EDITOR
     /// with a scaffold to create it.
-    #[arg(long, value_name = "NAME")]
+    #[arg(long, value_name = "NAME", help_heading = "Session & Memory")]
     pub skill: Vec<String>,
-    /// Input text
+
-    #[arg(trailing_var_arg = true)]
+    /// Start an agent
-    text: Vec<String>,
+    #[arg(short = 'a', long, help_heading = "Agents, RAG & Macros", add = ArgValueCompleter::new(agent_completer))]
-    /// Tail logs
+    pub agent: Option<String>,
-    #[arg(long)]
+    /// Set agent variables
-    pub tail_logs: bool,
+    #[arg(long, value_names = ["NAME", "VALUE"], num_args = 2, help_heading = "Agents, RAG & Macros")]
-    /// Disable colored log output
+    pub agent_variable: Vec<String>,
-    #[arg(long, requires = "tail_logs")]
+    /// Start a RAG
-    pub disable_log_colors: bool,
+    #[arg(long, help_heading = "Agents, RAG & Macros", add = ArgValueCompleter::new(rag_completer))]
-    /// Add a secret to the Coyote vault
+    pub rag: Option<String>,
-    #[arg(long, value_name = "SECRET_NAME", exclusive = true)]
+    /// Rebuild the RAG to sync document changes
-    pub add_secret: Option<String>,
+    #[arg(long, help_heading = "Agents, RAG & Macros")]
-    /// Decrypt a secret from the Coyote vault and print the plaintext
+    pub rebuild_rag: bool,
-    #[arg(long, value_name = "SECRET_NAME", exclusive = true, add = ArgValueCompleter::new(secrets_completer))]
+    /// Execute a macro
-    pub get_secret: Option<String>,
+    #[arg(long = "macro", value_name = "MACRO", help_heading = "Agents, RAG & Macros", add = ArgValueCompleter::new(macro_completer))]
-    /// Update an existing secret in the Coyote vault
+    pub macro_name: Option<String>,
-    #[arg(long, value_name = "SECRET_NAME", exclusive = true, add = ArgValueCompleter::new(secrets_completer))]
+
-    pub update_secret: Option<String>,
+    /// List all available chat models
-    /// Delete a secret from the Coyote vault
+    #[arg(long, help_heading = "List & Discovery")]
-    #[arg(long, value_name = "SECRET_NAME", exclusive = true, add = ArgValueCompleter::new(secrets_completer))]
+    pub list_models: bool,
-    pub delete_secret: Option<String>,
+    /// List all roles
-    /// List all secrets stored in the Coyote vault
+    #[arg(long, help_heading = "List & Discovery")]
-    #[arg(long, exclusive = true)]
+    pub list_roles: bool,
-    pub list_secrets: bool,
+    /// List all sessions
-    /// Authenticate with an LLM provider using OAuth (e.g., --authenticate client_name)
+    #[arg(long, help_heading = "List & Discovery")]
-    #[arg(long, exclusive = true, value_name = "CLIENT_NAME")]
+    pub list_sessions: bool,
-    pub authenticate: Option<Option<String>>,
+    /// List all agents
-    /// Generate static shell completion scripts
+    #[arg(long, help_heading = "List & Discovery")]
-    #[arg(long, value_name = "SHELL", value_enum)]
+    pub list_agents: bool,
-    pub completions: Option<ShellCompletion>,
+    /// List all RAGs
+    #[arg(long, help_heading = "List & Discovery")]
+    pub list_rags: bool,
+    /// List all macros
+    #[arg(long, help_heading = "List & Discovery")]
+    pub list_macros: bool,
+    /// List all installed skills
+    #[arg(long, help_heading = "List & Discovery")]
+    pub list_skills: bool,
+
+    /// Reinstall bundled assets, overwriting any local changes
+    #[arg(
+        long,
+        value_name = "CATEGORY",
+        value_enum,
+        help_heading = "Installation & Updates"
+    )]
+    pub install: Option<AssetCategory>,
+    /// Install assets from a remote git repository (URL may be suffixed with #<ref>)
+    #[arg(long, value_name = "GIT_URL", help_heading = "Installation & Updates")]
+    pub install_from: Option<String>,
+    /// Restrict --install-from to a single asset category
+    #[arg(
+        long,
+        value_name = "CATEGORY",
+        value_enum,
+        requires = "install_from",
+        help_heading = "Installation & Updates"
+    )]
+    pub filter: Option<InstallFilter>,
+    /// Overwrite all conflicts without prompting (used with --install-from)
+    #[arg(
+        long,
+        requires = "install_from",
+        help_heading = "Installation & Updates"
+    )]
+    pub install_force: bool,
+    /// Sync models updates
+    #[arg(long, help_heading = "Installation & Updates")]
+    pub sync_models: bool,
     /// Update Coyote to the latest release, or to a specific version
-    #[arg(long, value_name = "VERSION")]
+    #[arg(long, value_name = "VERSION", help_heading = "Installation & Updates")]
     pub update: Option<Option<String>>,
     /// With --update, update even if Coyote was installed via a package manager
-    #[arg(long, requires = "update")]
+    #[arg(long, requires = "update", help_heading = "Installation & Updates")]
     pub force: bool,
+
+    /// Add a secret to the Coyote vault
+    #[arg(
+        long,
+        value_name = "SECRET_NAME",
+        exclusive = true,
+        help_heading = "Vault & Secrets"
+    )]
+    pub add_secret: Option<String>,
+    /// Decrypt a secret from the Coyote vault and print the plaintext
+    #[arg(long, value_name = "SECRET_NAME", exclusive = true, help_heading = "Vault & Secrets", add = ArgValueCompleter::new(secrets_completer))]
+    pub get_secret: Option<String>,
+    /// Update an existing secret in the Coyote vault
+    #[arg(long, value_name = "SECRET_NAME", exclusive = true, help_heading = "Vault & Secrets", add = ArgValueCompleter::new(secrets_completer))]
+    pub update_secret: Option<String>,
+    /// Delete a secret from the Coyote vault
+    #[arg(long, value_name = "SECRET_NAME", exclusive = true, help_heading = "Vault & Secrets", add = ArgValueCompleter::new(secrets_completer))]
+    pub delete_secret: Option<String>,
+    /// List all secrets stored in the Coyote vault
+    #[arg(long, exclusive = true, help_heading = "Vault & Secrets")]
+    pub list_secrets: bool,
+
+    /// Authenticate with an LLM provider using OAuth (e.g., --authenticate client_name)
+    #[arg(
+        long,
+        exclusive = true,
+        value_name = "CLIENT_NAME",
+        help_heading = "Authentication"
+    )]
+    pub authenticate: Option<Option<String>>,
+    /// Authenticate with an OAuth-protected remote MCP server (e.g., --auth-mcp server_name)
+    #[arg(long, exclusive = true, value_name = "SERVER_NAME", help_heading = "Authentication", add = ArgValueCompleter::new(mcp_server_completer))]
+    pub auth_mcp: Option<String>,
+
+    /// Launch Coyote inside a Docker sandbox (via `sbx`); name defaults to current directory basename
+    #[arg(long, value_name = "NAME", help_heading = "Sandbox")]
+    pub sandbox: Option<Option<String>>,
+    /// Create the sandbox without bootstrapping the host config or vault password file
+    #[arg(long, requires = "sandbox", help_heading = "Sandbox")]
+    pub fresh: bool,
+    /// Skip discovery and application of all sbx mixins (user and built-in)
+    #[arg(long, requires = "sandbox", help_heading = "Sandbox")]
+    pub no_mixins: bool,
+
+    /// Display information
+    #[arg(long, help_heading = "Diagnostics & Tools")]
+    pub info: bool,
+    /// Build all configured Bash tool scripts
+    #[arg(long, help_heading = "Diagnostics & Tools")]
+    pub build_tools: bool,
+    /// Tail logs
+    #[arg(long, help_heading = "Diagnostics & Tools")]
+    pub tail_logs: bool,
+    /// Disable colored log output
+    #[arg(long, requires = "tail_logs", help_heading = "Diagnostics & Tools")]
+    pub disable_log_colors: bool,
+
+    /// Generate static shell completion scripts
+    #[arg(long, value_name = "SHELL", value_enum, help_heading = "Shell")]
+    pub completions: Option<ShellCompletion>,
 }
 
 impl Cli {
@@ -393,6 +482,18 @@ mod tests {
         assert!(cli.build_tools);
     }
 
+    #[test]
+    fn parse_dangerously_skip_permissions_flag() {
+        let cli = parse(&["--dangerously-skip-permissions"]);
+        assert!(cli.dangerously_skip_permissions);
+    }
+
+    #[test]
+    fn parse_dangerously_skip_permissions_default_off() {
+        let cli = parse(&[]);
+        assert!(!cli.dangerously_skip_permissions);
+    }
+
     #[test]
     fn parse_sync_models_flag() {
         let cli = parse(&["--sync-models"]);
@@ -489,4 +590,59 @@ mod tests {
     fn parse_force_without_update_fails() {
         assert!(Cli::try_parse_from(["coyote", "--force"]).is_err());
     }
+
+    #[test]
+    fn parse_sandbox_flag_no_value() {
+        let cli = parse(&["--sandbox"]);
+        assert_eq!(cli.sandbox, Some(None));
+    }
+
+    #[test]
+    fn parse_sandbox_flag_with_name() {
+        let cli = parse(&["--sandbox", "my-box"]);
+        assert_eq!(cli.sandbox, Some(Some("my-box".to_string())));
+    }
+
+    #[test]
+    fn parse_sandbox_is_exclusive() {
+        assert!(Cli::try_parse_from(["coyote", "--sandbox", "--agent", "foo"]).is_err());
+    }
+
+    #[test]
+    fn parse_fresh_flag_requires_sandbox() {
+        assert!(Cli::try_parse_from(["coyote", "--fresh"]).is_err());
+    }
+
+    #[test]
+    fn parse_fresh_flag_with_sandbox() {
+        let cli = parse(&["--sandbox", "--fresh"]);
+        assert_eq!(cli.sandbox, Some(None));
+        assert!(cli.fresh);
+    }
+
+    #[test]
+    fn parse_fresh_flag_with_named_sandbox() {
+        let cli = parse(&["--sandbox", "foo", "--fresh"]);
+        assert_eq!(cli.sandbox, Some(Some("foo".to_string())));
+        assert!(cli.fresh);
+    }
+
+    #[test]
+    fn parse_no_mixins_requires_sandbox() {
+        assert!(Cli::try_parse_from(["coyote", "--no-mixins"]).is_err());
+    }
+
+    #[test]
+    fn parse_no_mixins_with_sandbox() {
+        let cli = parse(&["--sandbox", "--no-mixins"]);
+        assert!(cli.no_mixins);
+    }
+
+    #[test]
+    fn parse_sandbox_with_fresh_and_no_mixins() {
+        let cli = parse(&["--sandbox", "foo", "--fresh", "--no-mixins"]);
+        assert_eq!(cli.sandbox, Some(Some("foo".to_string())));
+        assert!(cli.fresh);
+        assert!(cli.no_mixins);
+    }
 }

src/client/access_token.rs

@@ -4,29 +4,44 @@ use indexmap::IndexMap;
 use parking_lot::RwLock;
 use std::sync::LazyLock;
 
-static ACCESS_TOKENS: LazyLock<RwLock<IndexMap<String, (String, i64)>>> =
+type AccessTokenEntry = (String, i64, Option<String>);
+
+static ACCESS_TOKENS: LazyLock<RwLock<IndexMap<String, AccessTokenEntry>>> =
     LazyLock::new(|| RwLock::new(IndexMap::new()));
 
 pub fn get_access_token(client_name: &str) -> Result<String> {
     ACCESS_TOKENS
         .read()
         .get(client_name)
-        .map(|(token, _)| token.clone())
+        .map(|(token, _, _)| token.clone())
         .ok_or_else(|| anyhow!("Invalid access token"))
 }
 
+pub fn get_access_token_account_id(client_name: &str) -> Option<String> {
+    ACCESS_TOKENS
+        .read()
+        .get(client_name)
+        .and_then(|(_, _, account_id)| account_id.clone())
+}
+
 pub fn is_valid_access_token(client_name: &str) -> bool {
     let access_tokens = ACCESS_TOKENS.read();
-    let (token, expires_at) = match access_tokens.get(client_name) {
+    let (token, expires_at, _) = match access_tokens.get(client_name) {
         Some(v) => v,
         None => return false,
     };
     !token.is_empty() && Utc::now().timestamp() < *expires_at
 }
 
-pub fn set_access_token(client_name: &str, token: String, expires_at: i64) {
+pub fn set_access_token(
+    client_name: &str,
+    token: String,
+    expires_at: i64,
+    account_id: Option<String>,
+) {
     let mut access_tokens = ACCESS_TOKENS.write();
     let entry = access_tokens.entry(client_name.to_string()).or_default();
     entry.0 = token;
     entry.1 = expires_at;
+    entry.2 = account_id;
 }

src/client/azure_openai.rs

@@ -50,7 +50,7 @@ fn prepare_chat_completions(
 
     let url = format!(
         "{}/openai/deployments/{}/chat/completions?api-version=2024-12-01-preview",
-        &api_base,
+        api_base,
         self_.model.real_name()
     );
 
@@ -69,7 +69,7 @@ fn prepare_embeddings(self_: &AzureOpenAIClient, data: &EmbeddingsData) -> Resul
 
     let url = format!(
         "{}/openai/deployments/{}/embeddings?api-version=2024-10-21",
-        &api_base,
+        api_base,
         self_.model.real_name()
     );
 

src/client/bedrock.rs

@@ -325,6 +325,7 @@ fn build_chat_completions_body(data: ChatCompletionsData, model: &Model) -> Resu
         mut messages,
         temperature,
         top_p,
+        reasoning_effort,
         functions,
         stream: _,
     } = data;
@@ -396,6 +397,11 @@ fn build_chat_completions_body(data: ChatCompletionsData, model: &Model) -> Resu
                         }))
                     }
                     for tool_result in tool_results {
+                        if let Some(round_text) = &tool_result.text {
+                            assistant_parts.push(json!({
+                                "text": round_text,
+                            }))
+                        }
                         assistant_parts.push(json!({
                             "toolUse": {
                                 "toolUseId": tool_result.call.id,
@@ -457,6 +463,9 @@ fn build_chat_completions_body(data: ChatCompletionsData, model: &Model) -> Resu
     if let Some(v) = top_p {
         body["inferenceConfig"]["topP"] = v.into();
     }
+    if let Some(v) = reasoning_effort {
+        body["additionalModelRequestFields"] = json!({ "output_config": { "effort": v } });
+    }
     if let Some(functions) = functions {
         let tools: Vec<_> = functions
             .iter()
@@ -520,7 +529,7 @@ fn extract_chat_completions(data: &Value) -> Result<ChatCompletionsOutput> {
         bail!("Invalid response data: {data}");
     }
 
-    let output = ChatCompletionsOutput { text, tool_calls };
+    let output = ChatCompletionsOutput { text, tool_calls, ..Default::default() };
     Ok(output)
 }
 

src/client/claude.rs

@@ -168,12 +168,22 @@ pub async fn claude_chat_completions_streaming(
     let mut function_arguments = String::new();
     let mut function_id = String::new();
     let mut reasoning_state = 0;
+    let mut thinking_text = String::new();
+    let mut thinking_signature = String::new();
     let handle = |message: SseMessage| -> Result<bool> {
         let data: Value = serde_json::from_str(&message.data)?;
         debug!("stream-data: {data}");
         if let Some(typ) = data["type"].as_str() {
             match typ {
                 "content_block_start" => {
+                    if let (Some("redacted_thinking"), Some(redacted_data)) = (
+                        data["content_block"]["type"].as_str(),
+                        data["content_block"]["data"].as_str(),
+                    ) {
+                        handler.thinking_block(ThinkingBlock::RedactedThinking {
+                            data: redacted_data.to_string(),
+                        });
+                    }
                     if let (Some("tool_use"), Some(name), Some(id)) = (
                         data["content_block"]["type"].as_str(),
                         data["content_block"]["name"].as_str(),
@@ -206,7 +216,10 @@ pub async fn claude_chat_completions_streaming(
                             handler.text("<think>\n")?;
                             reasoning_state = 1;
                         }
+                        thinking_text.push_str(text);
                         handler.text(text)?;
+                    } else if let Some(signature) = data["delta"]["signature"].as_str() {
+                        thinking_signature.push_str(signature);
                     } else if let (true, Some(partial_json)) = (
                         !function_name.is_empty(),
                         data["delta"]["partial_json"].as_str(),
@@ -218,6 +231,10 @@ pub async fn claude_chat_completions_streaming(
                     if reasoning_state == 1 {
                         handler.text("\n</think>\n\n")?;
                         reasoning_state = 0;
+                        handler.thinking_block(ThinkingBlock::Thinking {
+                            thinking: std::mem::take(&mut thinking_text),
+                            signature: std::mem::take(&mut thinking_signature),
+                        });
                     }
                     if !function_name.is_empty() {
                         let arguments: Value = if function_arguments.is_empty() {
@@ -251,6 +268,7 @@ pub fn claude_build_chat_completions_body(
         mut messages,
         temperature,
         top_p,
+        reasoning_effort,
         functions,
         stream,
     } = data;
@@ -312,13 +330,25 @@ pub fn claude_build_chat_completions_body(
                 }) => {
                     let mut assistant_parts = vec![];
                     let mut user_parts = vec![];
-                    if !text.is_empty() {
+                    for (index, tool_result) in tool_results.iter().enumerate() {
+                        for block in &tool_result.thinking {
+                            assistant_parts.push(json!(block));
+                        }
+                        let round_text = if index == 0 && !text.is_empty() {
+                            Some(text.as_str())
+                        } else {
+                            tool_result.text.as_deref()
+                        };
+                        if let Some(round_text) = round_text {
+                            let round_text = strip_think_tag(round_text);
+                            let round_text = round_text.trim();
+                            if !round_text.is_empty() {
                                 assistant_parts.push(json!({
                                     "type": "text",
-                            "text": text,
+                                    "text": round_text,
                                 }))
                             }
-                    for tool_result in tool_results {
+                        }
                         assistant_parts.push(json!({
                             "type": "tool_use",
                             "id": tool_result.call.id,
@@ -369,6 +399,9 @@ pub fn claude_build_chat_completions_body(
     if let Some(v) = top_p {
         body["top_p"] = v.into();
     }
+    if let Some(v) = reasoning_effort {
+        body["output_config"] = json!({ "effort": v });
+    }
     if stream {
         body["stream"] = true.into();
     }
@@ -399,12 +432,24 @@ pub fn claude_extract_chat_completions(data: &Value) -> Result<ChatCompletionsOu
     let mut text = String::new();
     let mut reasoning = None;
     let mut tool_calls = vec![];
+    let mut thinking = vec![];
     if let Some(list) = data["content"].as_array() {
         for item in list {
             match item["type"].as_str() {
                 Some("thinking") => {
                     if let Some(v) = item["thinking"].as_str() {
                         reasoning = Some(v.to_string());
+                        thinking.push(ThinkingBlock::Thinking {
+                            thinking: v.to_string(),
+                            signature: item["signature"].as_str().unwrap_or_default().to_string(),
+                        });
+                    }
+                }
+                Some("redacted_thinking") => {
+                    if let Some(v) = item["data"].as_str() {
+                        thinking.push(ThinkingBlock::RedactedThinking {
+                            data: v.to_string(),
+                        });
                     }
                 }
                 Some("text") => {
@@ -443,6 +488,7 @@ pub fn claude_extract_chat_completions(data: &Value) -> Result<ChatCompletionsOu
     let output = ChatCompletionsOutput {
         text: text.to_string(),
         tool_calls,
+        thinking,
     };
     Ok(output)
 }

src/client/claude_oauth.rs

@@ -25,8 +25,8 @@ impl OAuthProvider for ClaudeOAuthProvider {
         "https://console.anthropic.com/oauth/code/callback"
     }
 
-    fn scopes(&self) -> &str {
+    fn scopes(&self) -> String {
-        "org:create_api_key user:profile user:inference"
+        "org:create_api_key user:profile user:inference".to_string()
     }
 
     fn extra_authorize_params(&self) -> Vec<(&str, &str)> {

src/client/cohere.rs

@@ -244,6 +244,6 @@ fn extract_chat_completions(data: &Value) -> Result<ChatCompletionsOutput> {
     if text.is_empty() && tool_calls.is_empty() {
         bail!("Invalid response data: {data}");
     }
-    let output = ChatCompletionsOutput { text, tool_calls };
+    let output = ChatCompletionsOutput { text, tool_calls, ..Default::default() };
     Ok(output)
 }

src/client/common.rs

@@ -286,6 +286,7 @@ pub struct ChatCompletionsData {
     pub messages: Vec<Message>,
     pub temperature: Option<f64>,
     pub top_p: Option<f64>,
+    pub reasoning_effort: Option<String>,
     pub functions: Option<Vec<FunctionDeclaration>>,
     pub stream: bool,
 }
@@ -294,6 +295,7 @@ pub struct ChatCompletionsData {
 pub struct ChatCompletionsOutput {
     pub text: String,
     pub tool_calls: Vec<ToolCall>,
+    pub thinking: Vec<ThinkingBlock>,
 }
 
 impl ChatCompletionsOutput {
@@ -401,10 +403,25 @@ pub async fn create_openai_compatible_client_config(
     };
     config["api_base"] = api_base.into();
 
+    let has_bundled_oauth = ALL_PROVIDER_MODELS
+        .iter()
+        .any(|p| p.provider == client && p.oauth.is_some());
+
+    let use_oauth = if has_bundled_oauth {
+        let choice = Select::new("Authentication method:", vec!["API Key", "OAuth"]).prompt()?;
+        choice == "OAuth"
+    } else {
+        false
+    };
+
+    if use_oauth {
+        config["auth"] = "oauth".into();
+    } else {
         let api_key = prompt_input_string("API Key", false, None)?;
         if !api_key.is_empty() {
             config["api_key"] = api_key.into();
         }
+    }
 
     let model = set_client_models_config(&mut config, &name).await?;
     let clients = json!(vec![config]);
@@ -434,6 +451,7 @@ pub async fn call_chat_completions(
             let ChatCompletionsOutput {
                 mut text,
                 tool_calls,
+                thinking,
                 ..
             } = ret;
             if !text.is_empty() {
@@ -444,7 +462,10 @@ pub async fn call_chat_completions(
                     ctx.app.config.print_markdown(&text)?;
                 }
             }
-            let tool_results = eval_tool_calls(ctx, tool_calls).await?;
+            let mut tool_results = eval_tool_calls(ctx, tool_calls).await?;
+            if let Some(first) = tool_results.first_mut() {
+                first.thinking = thinking;
+            }
             tool_results
                 .iter()
                 .for_each(|res| ctx.tool_scope.tool_tracker.record_call(res.call.clone()));
@@ -478,13 +499,16 @@ pub async fn call_chat_completions_streaming(
 
     render_ret?;
 
-    let (text, tool_calls) = handler.take();
+    let (text, tool_calls, thinking) = handler.take();
     match send_ret {
         Ok(_) => {
             if !text.is_empty() && !text.ends_with('\n') {
                 println!();
             }
-            let tool_results = eval_tool_calls(ctx, tool_calls).await?;
+            let mut tool_results = eval_tool_calls(ctx, tool_calls).await?;
+            if let Some(first) = tool_results.first_mut() {
+                first.thinking = thinking;
+            }
             tool_results
                 .iter()
                 .for_each(|res| ctx.tool_scope.tool_tracker.record_call(res.call.clone()));

src/client/gemini_oauth.rs

@@ -27,8 +27,8 @@ impl OAuthProvider for GeminiOAuthProvider {
         ""
     }
 
-    fn scopes(&self) -> &str {
+    fn scopes(&self) -> String {
-        "https://www.googleapis.com/auth/generative-language.peruserquota https://www.googleapis.com/auth/generative-language.retriever https://www.googleapis.com/auth/userinfo.email"
+        "https://www.googleapis.com/auth/generative-language.peruserquota https://www.googleapis.com/auth/generative-language.retriever https://www.googleapis.com/auth/userinfo.email".to_string()
     }
 
     fn client_secret(&self) -> Option<&str> {

src/client/message.rs

@@ -118,6 +118,9 @@ impl MessageContent {
                     lines.push(text.clone())
                 }
                 for tool_result in tool_results {
+                    if let Some(round_text) = &tool_result.text {
+                        lines.push(round_text.clone())
+                    }
                     let mut parts = vec!["Call".to_string()];
                     if let Some((agent_name, functions)) = agent_info
                         && functions.contains(&tool_result.call.name)
@@ -133,6 +136,13 @@ impl MessageContent {
         }
     }
 
+    pub fn as_text(&self) -> Option<&str> {
+        match self {
+            MessageContent::Text(text) => Some(text),
+            _ => None,
+        }
+    }
+
     pub fn merge_prompt(&mut self, replace_fn: impl Fn(&str) -> String) {
         match self {
             MessageContent::Text(text) => *text = replace_fn(text),
@@ -178,6 +188,17 @@ pub struct ImageUrl {
     pub url: String,
 }
 
+/// An extended-thinking block returned by Anthropic-protocol models.
+/// Serialized to match the API wire format (`type: thinking` / `type: redacted_thinking`)
+/// so blocks can be replayed verbatim, signature intact, in subsequent
+/// tool-loop rounds as the API requires.
+#[derive(Debug, Clone, Deserialize, Serialize)]
+#[serde(tag = "type", rename_all = "snake_case")]
+pub enum ThinkingBlock {
+    Thinking { thinking: String, signature: String },
+    RedactedThinking { data: String },
+}
+
 #[derive(Debug, Clone, Deserialize, Serialize)]
 pub struct MessageContentToolCalls {
     pub tool_results: Vec<ToolResult>,
@@ -194,9 +215,13 @@ impl MessageContentToolCalls {
         }
     }
 
-    pub fn merge(&mut self, tool_results: Vec<ToolResult>, _text: String) {
+    pub fn merge(&mut self, mut tool_results: Vec<ToolResult>, text: String) {
+        if !text.is_empty()
+            && let Some(first) = tool_results.first_mut()
+        {
+            first.text = Some(text);
+        }
         self.tool_results.extend(tool_results);
-        self.text.clear();
         self.sequence = true;
     }
 }

src/client/mod.rs

@@ -4,6 +4,8 @@ mod common;
 mod gemini_oauth;
 mod message;
 pub mod oauth;
+mod openai_compatible_oauth;
+mod openai_oauth;
 #[macro_use]
 mod macros;
 mod model;

src/client/model.rs

@@ -6,6 +6,7 @@ use super::{
 use crate::config::AppConfig;
 use crate::utils::{estimate_token_length, strip_think_tag};
 
+use super::oauth::OAuthConfig;
 use anyhow::{Result, bail};
 use serde::{Deserialize, Serialize};
 use serde_json::Value;
@@ -289,6 +290,14 @@ impl Model {
         }
         Ok(())
     }
+
+    pub fn reasoning_levels(&self) -> &[String] {
+        &self.data.reasoning_levels
+    }
+
+    pub fn default_reasoning_effort(&self) -> Option<&str> {
+        self.data.default_reasoning_effort.as_deref()
+    }
 }
 
 #[derive(Debug, Clone, Default, Serialize, Deserialize)]
@@ -316,6 +325,10 @@ pub struct ModelData {
     pub supports_vision: bool,
     #[serde(default, skip_serializing_if = "std::ops::Not::not")]
     pub supports_function_calling: bool,
+    #[serde(default, skip_serializing_if = "Vec::is_empty")]
+    pub reasoning_levels: Vec<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub default_reasoning_effort: Option<String>,
     #[serde(default, skip_serializing_if = "std::ops::Not::not")]
     no_stream: bool,
     #[serde(default, skip_serializing_if = "std::ops::Not::not")]
@@ -345,6 +358,8 @@ impl ModelData {
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct ProviderModels {
     pub provider: String,
+    #[serde(default)]
+    pub oauth: Option<OAuthConfig>,
     pub models: Vec<ModelData>,
 }
 

src/client/openai.rs

@@ -1,13 +1,17 @@
+use super::access_token::{get_access_token, get_access_token_account_id};
+use super::oauth::{self, OAuthProvider};
+use super::openai_oauth::OpenAIOAuthProvider;
 use super::*;
 
 use crate::utils::strip_think_tag;
 
 use anyhow::{Context, Result, bail};
-use reqwest::RequestBuilder;
+use reqwest::{Client as ReqwestClient, RequestBuilder};
 use serde::Deserialize;
 use serde_json::{Value, json};
 
 const API_BASE: &str = "https://api.openai.com/v1";
+const CODEX_API_ENDPOINT: &str = "https://chatgpt.com/backend-api/codex/responses";
 
 #[derive(Debug, Clone, Deserialize, Default)]
 pub struct OpenAIConfig {
@@ -15,6 +19,7 @@ pub struct OpenAIConfig {
     pub api_key: Option<String>,
     pub api_base: Option<String>,
     pub organization_id: Option<String>,
+    pub auth: Option<String>,
     #[serde(default)]
     pub models: Vec<ModelData>,
     pub patch: Option<RequestPatch>,
@@ -25,36 +30,131 @@ impl OpenAIClient {
     config_get_fn!(api_key, get_api_key);
     config_get_fn!(api_base, get_api_base);
 
-    create_client_config!([("api_key", "API Key", None, true)]);
+    create_oauth_supported_client_config!();
 }
 
-impl_client_trait!(
+#[async_trait::async_trait]
-    OpenAIClient,
+impl Client for OpenAIClient {
-    (
+    client_common_fns!();
-        prepare_chat_completions,
-        openai_chat_completions,
-        openai_chat_completions_streaming
-    ),
-    (prepare_embeddings, openai_embeddings),
-    (noop_prepare_rerank, noop_rerank),
-);
 
-fn prepare_chat_completions(
+    fn supports_oauth(&self) -> bool {
+        self.config.auth.as_deref() == Some("oauth")
+    }
+
+    async fn chat_completions_inner(
+        &self,
+        client: &ReqwestClient,
+        data: ChatCompletionsData,
+    ) -> Result<ChatCompletionsOutput> {
+        let uses_codex =
+            self.config.auth.as_deref() == Some("oauth") && self.get_api_base().is_err();
+        let request_data = prepare_chat_completions(self, client, data).await?;
+        let builder = self.request_builder(client, request_data);
+        if uses_codex {
+            openai_responses_chat_completions(builder, self.model()).await
+        } else {
+            openai_chat_completions(builder, self.model()).await
+        }
+    }
+
+    async fn chat_completions_streaming_inner(
+        &self,
+        client: &ReqwestClient,
+        handler: &mut SseHandler,
+        data: ChatCompletionsData,
+    ) -> Result<()> {
+        let uses_codex =
+            self.config.auth.as_deref() == Some("oauth") && self.get_api_base().is_err();
+        let request_data = prepare_chat_completions(self, client, data).await?;
+        let builder = self.request_builder(client, request_data);
+
+        if uses_codex {
+            openai_responses_streaming(builder, handler).await
+        } else {
+            openai_chat_completions_streaming(builder, handler, self.model()).await
+        }
+    }
+
+    async fn embeddings_inner(
+        &self,
+        client: &ReqwestClient,
+        data: &EmbeddingsData,
+    ) -> Result<EmbeddingsOutput> {
+        let request_data = prepare_embeddings(self, client, data).await?;
+        let builder = self.request_builder(client, request_data);
+        openai_embeddings(builder, self.model()).await
+    }
+
+    async fn rerank_inner(
+        &self,
+        client: &ReqwestClient,
+        data: &RerankData,
+    ) -> Result<RerankOutput> {
+        let request_data = noop_prepare_rerank(self, data)?;
+        let builder = self.request_builder(client, request_data);
+        noop_rerank(builder, self.model()).await
+    }
+}
+
+async fn prepare_chat_completions(
     self_: &OpenAIClient,
+    client: &ReqwestClient,
     data: ChatCompletionsData,
 ) -> Result<RequestData> {
-    let api_key = self_.get_api_key()?;
+    let uses_oauth = self_.config.auth.as_deref() == Some("oauth");
+    let has_custom_base = self_.get_api_base().is_ok();
+
+    let uses_codex = uses_oauth && !has_custom_base;
+
+    let url = if uses_codex {
+        CODEX_API_ENDPOINT.to_string()
+    } else {
         let api_base = self_
             .get_api_base()
             .unwrap_or_else(|_| API_BASE.to_string());
+        format!("{}/chat/completions", api_base.trim_end_matches('/'))
+    };
 
-    let url = format!("{}/chat/completions", api_base.trim_end_matches('/'));
+    let body = if uses_codex {
-
+        openai_build_responses_body(data, &self_.model)
-    let body = openai_build_chat_completions_body(data, &self_.model);
+    } else {
+        openai_build_chat_completions_body(data, &self_.model)
+    };
 
     let mut request_data = RequestData::new(url, body);
 
+    if uses_oauth {
+        let provider = OpenAIOAuthProvider;
+        let ready = oauth::prepare_oauth_access_token(client, &provider, self_.name()).await?;
+
+        if !ready {
+            bail!(
+                "OAuth configured but no tokens found for '{}'. Run: 'coyote --authenticate {}' or '.authenticate' in the REPL",
+                self_.name(),
+                self_.name()
+            );
+        }
+
+        let token = get_access_token(self_.name())?;
+        request_data.bearer_auth(token);
+
+        if let Some(account_id) = get_access_token_account_id(self_.name()) {
+            request_data.header("ChatGPT-Account-Id", account_id);
+        }
+
+        for (key, value) in provider.extra_request_headers() {
+            request_data.header(key, value);
+        }
+    } else if let Ok(api_key) = self_.get_api_key() {
         request_data.bearer_auth(api_key);
+    } else {
+        bail!(
+            "No authentication configured for '{}'. Set `api_key` or use `auth: oauth` with `coyote --authenticate {}`.",
+            self_.name(),
+            self_.name()
+        );
+    }
+
     if let Some(organization_id) = &self_.config.organization_id {
         request_data.header("OpenAI-Organization", organization_id);
     }
@@ -62,8 +162,11 @@ fn prepare_chat_completions(
     Ok(request_data)
 }
 
-fn prepare_embeddings(self_: &OpenAIClient, data: &EmbeddingsData) -> Result<RequestData> {
+async fn prepare_embeddings(
-    let api_key = self_.get_api_key()?;
+    self_: &OpenAIClient,
+    client: &ReqwestClient,
+    data: &EmbeddingsData,
+) -> Result<RequestData> {
     let api_base = self_
         .get_api_base()
         .unwrap_or_else(|_| API_BASE.to_string());
@@ -74,7 +177,30 @@ fn prepare_embeddings(self_: &OpenAIClient, data: &EmbeddingsData) -> Result<Req
 
     let mut request_data = RequestData::new(url, body);
 
+    if self_.config.auth.as_deref() == Some("oauth") {
+        let provider = OpenAIOAuthProvider;
+        let ready = oauth::prepare_oauth_access_token(client, &provider, self_.name()).await?;
+
+        if !ready {
+            bail!(
+                "OAuth configured but no tokens found for '{}'. Run: 'coyote --authenticate {}' or '.authenticate' in the REPL",
+                self_.name(),
+                self_.name()
+            );
+        }
+
+        let token = get_access_token(self_.name())?;
+        request_data.bearer_auth(token);
+    } else if let Ok(api_key) = self_.get_api_key() {
         request_data.bearer_auth(api_key);
+    } else {
+        bail!(
+            "No authentication configured for '{}'. Set `api_key` or use `auth: oauth` with `coyote --authenticate {}`.",
+            self_.name(),
+            self_.name()
+        );
+    }
+
     if let Some(organization_id) = &self_.config.organization_id {
         request_data.header("OpenAI-Organization", organization_id);
     }
@@ -230,6 +356,7 @@ pub fn openai_build_chat_completions_body(data: ChatCompletionsData, model: &Mod
         messages,
         temperature,
         top_p,
+        reasoning_effort,
         functions,
         stream,
     } = data;
@@ -243,7 +370,7 @@ pub fn openai_build_chat_completions_body(data: ChatCompletionsData, model: &Mod
             match content {
                 MessageContent::ToolCalls(MessageContentToolCalls {
                     tool_results,
-                    text: _,
+                    text,
                     sequence,
                 }) => {
                     if !sequence {
@@ -260,9 +387,12 @@ pub fn openai_build_chat_completions_body(data: ChatCompletionsData, model: &Mod
                                 })
                             })
                             .collect();
-                        let mut messages = vec![
+                        let mut assistant_message =
-                            json!({ "role": MessageRole::Assistant, "tool_calls": tool_calls }),
+                            json!({ "role": MessageRole::Assistant, "tool_calls": tool_calls });
-                        ];
+                        if !text.is_empty() {
+                            assistant_message["content"] = strip_think_tag(&text).into();
+                        }
+                        let mut messages = vec![assistant_message];
                         for tool_result in tool_results {
                             messages.push(json!({
                                 "role": "tool",
@@ -272,9 +402,13 @@ pub fn openai_build_chat_completions_body(data: ChatCompletionsData, model: &Mod
                         }
                         messages
                     } else {
-                        tool_results.into_iter().flat_map(|tool_result| {
+                        tool_results.into_iter().enumerate().flat_map(|(index, tool_result)| {
-                            vec![
+                            let round_text = if index == 0 && !text.is_empty() {
-                                json!({
+                                Some(text.clone())
+                            } else {
+                                tool_result.text.clone()
+                            };
+                            let mut assistant_message = json!({
                                 "role": MessageRole::Assistant,
                                 "tool_calls": [
                                     {
@@ -286,7 +420,12 @@ pub fn openai_build_chat_completions_body(data: ChatCompletionsData, model: &Mod
                                         },
                                     }
                                 ]
-                                }),
+                            });
+                            if let Some(round_text) = round_text {
+                                assistant_message["content"] = strip_think_tag(&round_text).into();
+                            }
+                            vec![
+                                assistant_message,
                                 json!({
                                     "role": "tool",
                                     "content": tool_result.output.to_string(),
@@ -328,6 +467,9 @@ pub fn openai_build_chat_completions_body(data: ChatCompletionsData, model: &Mod
     if let Some(v) = top_p {
         body["top_p"] = v.into();
     }
+    if let Some(v) = reasoning_effort {
+        body["reasoning_effort"] = v.into();
+    }
     if stream {
         body["stream"] = true.into();
     }
@@ -391,7 +533,7 @@ pub fn openai_extract_chat_completions(data: &Value) -> Result<ChatCompletionsOu
     } else {
         text.to_string()
     };
-    let output = ChatCompletionsOutput { text, tool_calls };
+    let output = ChatCompletionsOutput { text, tool_calls, ..Default::default() };
     Ok(output)
 }
 
@@ -402,3 +544,204 @@ fn normalize_function_id(value: &str) -> Option<String> {
         Some(value.to_string())
     }
 }
+
+pub fn openai_build_responses_body(data: ChatCompletionsData, model: &Model) -> Value {
+    let ChatCompletionsData {
+        messages,
+        temperature,
+        top_p,
+        reasoning_effort,
+        functions,
+        stream,
+    } = data;
+
+    let messages_len = messages.len();
+    let input: Vec<Value> = messages
+        .into_iter()
+        .enumerate()
+        .flat_map(|(i, message)| {
+            let Message { role, content } = message;
+            match content {
+                MessageContent::ToolCalls(MessageContentToolCalls {
+                    tool_results,
+                    text,
+                    sequence: _,
+                }) => tool_results
+                    .into_iter()
+                    .enumerate()
+                    .flat_map(|(index, tool_result)| {
+                        let round_text = if index == 0 && !text.is_empty() {
+                            Some(text.clone())
+                        } else {
+                            tool_result.text.clone()
+                        };
+                        let mut items = vec![];
+                        if let Some(round_text) = round_text {
+                            items.push(json!({
+                                "role": MessageRole::Assistant,
+                                "content": strip_think_tag(&round_text),
+                            }));
+                        }
+                        items.push(json!({
+                            "type": "function_call",
+                            "call_id": tool_result.call.id,
+                            "name": tool_result.call.name,
+                            "arguments": tool_result.call.arguments.to_string(),
+                        }));
+                        items.push(json!({
+                            "type": "function_call_output",
+                            "call_id": tool_result.call.id,
+                            "output": tool_result.output.to_string(),
+                        }));
+                        items
+                    })
+                    .collect(),
+                MessageContent::Text(text) if role.is_assistant() && i != messages_len - 1 => {
+                    vec![json!({ "role": role, "content": strip_think_tag(&text) })]
+                }
+                _ => vec![json!({ "role": role, "content": content })],
+            }
+        })
+        .collect();
+
+    let mut body = json!({
+        "model": &model.real_name(),
+        "input": input,
+        "store": false,
+    });
+
+    if let Some(v) = model.max_tokens_param() {
+        body["max_output_tokens"] = v.into();
+    }
+    if let Some(v) = temperature {
+        body["temperature"] = v.into();
+    }
+    if let Some(v) = top_p {
+        body["top_p"] = v.into();
+    }
+    if let Some(v) = reasoning_effort {
+        body["reasoning"] = json!({ "effort": v });
+    }
+    if stream {
+        body["stream"] = true.into();
+    }
+    if let Some(functions) = functions {
+        body["tools"] = functions
+            .iter()
+            .map(|v| {
+                let mut tool = serde_json::to_value(v).unwrap_or_default();
+                tool["type"] = "function".into();
+                tool
+            })
+            .collect();
+    }
+    body
+}
+
+pub async fn openai_responses_chat_completions(
+    builder: RequestBuilder,
+    _model: &Model,
+) -> Result<ChatCompletionsOutput> {
+    let res = builder.send().await?;
+    let status = res.status();
+    let data: Value = res.json().await?;
+
+    if !status.is_success() {
+        catch_error(&data, status.as_u16())?;
+    }
+
+    debug!("non-stream-data: {data}");
+    openai_extract_responses(&data)
+}
+
+pub fn openai_extract_responses(data: &Value) -> Result<ChatCompletionsOutput> {
+    let mut text = String::new();
+    let mut tool_calls = vec![];
+
+    if let Some(output) = data["output"].as_array() {
+        for item in output {
+            match item["type"].as_str() {
+                Some("message") => {
+                    if let Some(content) = item["content"].as_array() {
+                        for part in content {
+                            if part["type"].as_str() == Some("output_text")
+                                && let Some(t) = part["text"].as_str()
+                            {
+                                text.push_str(t);
+                            }
+                        }
+                    }
+                }
+                Some("function_call") => {
+                    if let (Some(name), Some(arguments_str), Some(call_id)) = (
+                        item["name"].as_str(),
+                        item["arguments"].as_str(),
+                        item["call_id"].as_str(),
+                    ) {
+                        let arguments: Value = arguments_str.parse().with_context(|| {
+                            format!("Tool call '{name}' has non-JSON arguments '{arguments_str}'")
+                        })?;
+                        tool_calls.push(ToolCall::new(
+                            name.to_string(),
+                            arguments,
+                            Some(call_id.to_string()),
+                        ));
+                    }
+                }
+                _ => {}
+            }
+        }
+    }
+
+    if text.is_empty() && tool_calls.is_empty() {
+        bail!("Invalid response data: {data}");
+    }
+    Ok(ChatCompletionsOutput { text, tool_calls, ..Default::default() })
+}
+
+pub async fn openai_responses_streaming(
+    builder: RequestBuilder,
+    handler: &mut SseHandler,
+) -> Result<()> {
+    let handle = |message: SseMessage| -> Result<bool> {
+        if message.data == "[DONE]" {
+            return Ok(true);
+        }
+        let data: Value = serde_json::from_str(&message.data)?;
+        debug!("stream-data: {data}");
+
+        match data["type"].as_str() {
+            Some("response.output_text.delta") => {
+                if let Some(delta) = data["delta"].as_str().filter(|v| !v.is_empty()) {
+                    handler.text(delta)?;
+                }
+            }
+            Some("response.output_item.done") => {
+                let item = &data["item"];
+                if item["type"].as_str() == Some("function_call")
+                    && let (Some(name), Some(arguments_str), Some(call_id)) = (
+                        item["name"].as_str(),
+                        item["arguments"].as_str(),
+                        item["call_id"].as_str(),
+                    )
+                {
+                    let arguments: Value = arguments_str.parse().with_context(|| {
+                        format!("Tool call '{name}' has non-JSON arguments '{arguments_str}'")
+                    })?;
+                    handler.tool_call(ToolCall::new(
+                        name.to_string(),
+                        arguments,
+                        Some(call_id.to_string()),
+                    ))?;
+                }
+            }
+            Some("response.completed") => {
+                return Ok(true);
+            }
+            _ => {}
+        }
+        Ok(false)
+    };
+
+    sse_stream(builder, handle).await
+}

src/client/openai_compatible.rs

@@ -1,16 +1,21 @@
+use super::access_token::get_access_token;
+use super::oauth;
 use super::openai::*;
 use super::*;
 
-use anyhow::{Context, Result};
+use anyhow::{Context, Result, anyhow, bail};
-use reqwest::RequestBuilder;
+use reqwest::{Client as ReqwestClient, RequestBuilder};
 use serde::Deserialize;
 use serde_json::{Value, json};
+use oauth::OAuthConfig;
 
 #[derive(Debug, Clone, Deserialize)]
 pub struct OpenAICompatibleConfig {
     pub name: Option<String>,
     pub api_base: Option<String>,
     pub api_key: Option<String>,
+    pub auth: Option<String>,
+    pub oauth: Option<Box<OAuthConfig>>,
     #[serde(default)]
     pub models: Vec<ModelData>,
     pub patch: Option<RequestPatch>,
@@ -24,78 +29,159 @@ impl OpenAICompatibleClient {
     create_client_config!([]);
 }
 
-impl_client_trait!(
+#[async_trait::async_trait]
-    OpenAICompatibleClient,
+impl Client for OpenAICompatibleClient {
-    (
+    client_common_fns!();
-        prepare_chat_completions,
-        openai_chat_completions,
-        openai_chat_completions_streaming
-    ),
-    (prepare_embeddings, openai_embeddings),
-    (prepare_rerank, generic_rerank),
-);
 
-fn prepare_chat_completions(
+    fn supports_oauth(&self) -> bool {
+        self.config.auth.as_deref() == Some("oauth")
+    }
+
+    async fn chat_completions_inner(
+        &self,
+        client: &ReqwestClient,
+        data: ChatCompletionsData,
+    ) -> Result<ChatCompletionsOutput> {
+        let request_data = prepare_chat_completions(self, client, data).await?;
+        let builder = self.request_builder(client, request_data);
+        
+        openai_chat_completions(builder, self.model()).await
+    }
+
+    async fn chat_completions_streaming_inner(
+        &self,
+        client: &ReqwestClient,
+        handler: &mut SseHandler,
+        data: ChatCompletionsData,
+    ) -> Result<()> {
+        let request_data = prepare_chat_completions(self, client, data).await?;
+        let builder = self.request_builder(client, request_data);
+        
+        openai_chat_completions_streaming(builder, handler, self.model()).await
+    }
+
+    async fn embeddings_inner(
+        &self,
+        client: &ReqwestClient,
+        data: &EmbeddingsData,
+    ) -> Result<EmbeddingsOutput> {
+        let request_data = prepare_embeddings(self, client, data).await?;
+        let builder = self.request_builder(client, request_data);
+        
+        openai_embeddings(builder, self.model()).await
+    }
+
+    async fn rerank_inner(
+        &self,
+        client: &ReqwestClient,
+        data: &RerankData,
+    ) -> Result<RerankOutput> {
+        let request_data = prepare_rerank(self, client, data).await?;
+        let builder = self.request_builder(client, request_data);
+        
+        generic_rerank(builder, self.model()).await
+    }
+}
+
+async fn prepare_chat_completions(
     self_: &OpenAICompatibleClient,
+    client: &ReqwestClient,
     data: ChatCompletionsData,
 ) -> Result<RequestData> {
-    let api_key = self_.get_api_key().ok();
     let api_base = get_api_base_ext(self_)?;
-
     let url = format!("{api_base}/chat/completions");
-
     let body = openai_build_chat_completions_body(data, &self_.model);
-
     let mut request_data = RequestData::new(url, body);
     
-    if let Some(api_key) = api_key {
+    apply_auth(self_, client, &mut request_data).await?;
-        request_data.bearer_auth(api_key);
-    }
     
     Ok(request_data)
 }
 
-fn prepare_embeddings(
+async fn prepare_embeddings(
     self_: &OpenAICompatibleClient,
+    client: &ReqwestClient,
     data: &EmbeddingsData,
 ) -> Result<RequestData> {
-    let api_key = self_.get_api_key().ok();
     let api_base = get_api_base_ext(self_)?;
-
     let url = format!("{api_base}/embeddings");
-
     let body = openai_build_embeddings_body(data, &self_.model);
-
     let mut request_data = RequestData::new(url, body);
     
-    if let Some(api_key) = api_key {
+    apply_auth(self_, client, &mut request_data).await?;
-        request_data.bearer_auth(api_key);
-    }
     
     Ok(request_data)
 }
 
-fn prepare_rerank(self_: &OpenAICompatibleClient, data: &RerankData) -> Result<RequestData> {
+async fn prepare_rerank(
-    let api_key = self_.get_api_key().ok();
+    self_: &OpenAICompatibleClient,
+    client: &ReqwestClient,
+    data: &RerankData,
+) -> Result<RequestData> {
     let api_base = get_api_base_ext(self_)?;
-
     let url = if self_.name().starts_with("ernie") {
         format!("{api_base}/rerankers")
     } else {
         format!("{api_base}/rerank")
     };
-
     let body = generic_build_rerank_body(data, &self_.model);
-
     let mut request_data = RequestData::new(url, body);
     
-    if let Some(api_key) = api_key {
+    apply_auth(self_, client, &mut request_data).await?;
-        request_data.bearer_auth(api_key);
-    }
     
     Ok(request_data)
 }
 
+async fn apply_auth(
+    self_: &OpenAICompatibleClient,
+    client: &ReqwestClient,
+    request_data: &mut RequestData,
+) -> Result<()> {
+    if self_.config.auth.as_deref() == Some("oauth") {
+        let client_name = self_.name();
+        let app_config = self_.app_config();
+        let cc = app_config
+            .clients
+            .iter()
+            .find(|cc| {
+                matches!(
+                    cc,
+                    ClientConfig::OpenAICompatibleConfig(c)
+                    if c.name.as_deref().unwrap_or("openai-compatible") == client_name
+                )
+            })
+            .ok_or_else(|| {
+                anyhow!("Could not locate ClientConfig entry for '{}'", client_name)
+            })?;
+        let provider = oauth::get_oauth_provider_for_client(cc, &ALL_PROVIDER_MODELS)
+            .ok_or_else(|| {
+                anyhow!(
+                    "OAuth configured for '{}' but no oauth block resolved (missing from both models.yaml and user config)",
+                    client_name
+                )
+            })?;
+        
+        let ready = oauth::prepare_oauth_access_token(client, &*provider, client_name).await?;
+        if !ready {
+            bail!(
+                "OAuth configured for '{}' but no tokens found. Run: 'coyote --authenticate {}' or '.authenticate' in the REPL",
+                client_name,
+                client_name
+            );
+        }
+        
+        let token = get_access_token(client_name)?;
+        request_data.bearer_auth(token);
+        
+        for (key, value) in provider.extra_request_headers() {
+            request_data.header(key, value);
+        }
+    } else if let Ok(api_key) = self_.get_api_key() {
+        request_data.bearer_auth(api_key);
+    }
+    Ok(())
+}
+
 fn get_api_base_ext(self_: &OpenAICompatibleClient) -> Result<String> {
     let api_base = match self_.get_api_base() {
         Ok(v) => v,

src/client/openai_compatible_oauth.rs

@@ -0,0 +1,113 @@
+use url::Url;
+
+use super::oauth::{OAuthConfig, OAuthFlow, OAuthProvider, TokenRequestFormat};
+
+pub struct OpenAICompatibleOAuthProvider {
+    pub config: OAuthConfig,
+    pub client_name: String,
+}
+
+fn is_loopback_uri(uri: &str) -> bool {
+    Url::parse(uri)
+        .ok()
+        .and_then(|u| u.host_str().map(str::to_string))
+        .is_some_and(|host| matches!(host.as_str(), "127.0.0.1" | "localhost" | "[::1]" | "::1"))
+}
+
+impl OAuthProvider for OpenAICompatibleOAuthProvider {
+    fn provider_name(&self) -> &str {
+        &self.client_name
+    }
+
+    fn client_id(&self) -> &str {
+        &self.config.client_id
+    }
+
+    fn authorize_url(&self) -> &str {
+        self.config.authorize_url.as_deref().unwrap_or("")
+    }
+
+    fn token_url(&self) -> &str {
+        &self.config.token_url
+    }
+
+    fn redirect_uri(&self) -> &str {
+        self.config.redirect_uri.as_deref().unwrap_or("")
+    }
+
+    fn scopes(&self) -> String {
+        self.config.scopes.join(" ")
+    }
+
+    fn client_secret(&self) -> Option<&str> {
+        self.config.client_secret.as_deref()
+    }
+
+    fn extra_authorize_params(&self) -> Vec<(&str, &str)> {
+        self.config
+            .extra_authorize_params
+            .iter()
+            .map(|(k, v)| (k.as_str(), v.as_str()))
+            .collect()
+    }
+
+    fn token_request_format(&self) -> TokenRequestFormat {
+        self.config
+            .token_request_format
+            .unwrap_or(TokenRequestFormat::FormUrlEncoded)
+    }
+
+    fn uses_localhost_redirect(&self) -> bool {
+        self.config.redirect_uri.is_none() && self.config.redirect_port.is_none()
+    }
+
+    fn extra_token_headers(&self) -> Vec<(&str, &str)> {
+        self.config
+            .extra_token_headers
+            .iter()
+            .map(|(k, v)| (k.as_str(), v.as_str()))
+            .collect()
+    }
+
+    fn extra_request_headers(&self) -> Vec<(&str, &str)> {
+        self.config
+            .extra_request_headers
+            .iter()
+            .map(|(k, v)| (k.as_str(), v.as_str()))
+            .collect()
+    }
+
+    fn fixed_redirect_uri(&self) -> Option<String> {
+        if let Some(uri) = &self.config.redirect_uri {
+            return if is_loopback_uri(uri) {
+                Some(uri.clone())
+            } else {
+                None
+            };
+        }
+        if let Some(port) = self.config.redirect_port {
+            return Some(format!("http://127.0.0.1:{port}/callback"));
+        }
+        None
+    }
+
+    fn include_state_in_token_exchange(&self) -> bool {
+        self.config.include_state_in_token_exchange
+    }
+
+    fn flow(&self) -> OAuthFlow {
+        self.config.flow
+    }
+
+    fn echo_pkce_in_token_exchange(&self) -> bool {
+        self.config.echo_pkce_in_token_exchange
+    }
+
+    fn device_authorization_url(&self) -> Option<&str> {
+        self.config.device_authorization_url.as_deref()
+    }
+
+    fn use_pkce_in_device_flow(&self) -> bool {
+        self.config.use_pkce_in_device_flow
+    }
+}

src/client/openai_oauth.rs

@@ -0,0 +1,79 @@
+use super::oauth::{OAuthProvider, TokenRequestFormat};
+use base64::Engine;
+use base64::engine::general_purpose::URL_SAFE_NO_PAD;
+use serde_json::Value;
+
+pub struct OpenAIOAuthProvider;
+
+impl OAuthProvider for OpenAIOAuthProvider {
+    fn provider_name(&self) -> &str {
+        "openai"
+    }
+
+    fn client_id(&self) -> &str {
+        "app_EMoamEEZ73f0CkXaXp7hrann"
+    }
+
+    fn authorize_url(&self) -> &str {
+        "https://auth.openai.com/oauth/authorize"
+    }
+
+    fn token_url(&self) -> &str {
+        "https://auth.openai.com/oauth/token"
+    }
+
+    fn redirect_uri(&self) -> &str {
+        "http://localhost:1455/auth/callback"
+    }
+
+    fn scopes(&self) -> String {
+        "openid profile email offline_access".to_string()
+    }
+
+    fn token_request_format(&self) -> TokenRequestFormat {
+        TokenRequestFormat::FormUrlEncoded
+    }
+
+    fn extra_authorize_params(&self) -> Vec<(&str, &str)> {
+        vec![
+            ("id_token_add_organizations", "true"),
+            ("codex_cli_simplified_flow", "true"),
+        ]
+    }
+
+    fn fixed_redirect_uri(&self) -> Option<String> {
+        Some("http://localhost:1455/auth/callback".to_string())
+    }
+
+    fn include_state_in_token_exchange(&self) -> bool {
+        false
+    }
+
+    fn extract_account_id(&self, response: &Value) -> Option<String> {
+        let id_token = response["id_token"].as_str().unwrap_or_default();
+        let access_token = response["access_token"].as_str().unwrap_or_default();
+        extract_account_id_from_jwt(id_token).or_else(|| extract_account_id_from_jwt(access_token))
+    }
+}
+
+fn extract_account_id_from_jwt(token: &str) -> Option<String> {
+    let parts: Vec<&str> = token.splitn(3, '.').collect();
+    if parts.len() < 2 {
+        return None;
+    }
+    let decoded = URL_SAFE_NO_PAD.decode(parts[1]).ok()?;
+    let claims: Value = serde_json::from_slice(&decoded).ok()?;
+    claims["chatgpt_account_id"]
+        .as_str()
+        .map(|s| s.to_string())
+        .or_else(|| {
+            claims["https://api.openai.com/auth"]["chatgpt_account_id"]
+                .as_str()
+                .map(|s| s.to_string())
+        })
+        .or_else(|| {
+            claims["organizations"][0]["id"]
+                .as_str()
+                .map(|s| s.to_string())
+        })
+}

src/client/stream.rs

@@ -1,4 +1,4 @@
-use super::{ToolCall, catch_error};
+use super::{ThinkingBlock, ToolCall, catch_error};
 use crate::utils::AbortSignal;
 
 use anyhow::{Context, Result, anyhow, bail};
@@ -13,6 +13,7 @@ pub struct SseHandler {
     abort_signal: AbortSignal,
     buffer: String,
     tool_calls: Vec<ToolCall>,
+    thinking: Vec<ThinkingBlock>,
     last_tool_calls: Vec<ToolCall>,
     max_call_repeats: usize,
     call_repeat_chain_len: usize,
@@ -26,6 +27,7 @@ impl SseHandler {
             abort_signal,
             buffer: String::new(),
             tool_calls: Vec::new(),
+            thinking: Vec::new(),
             last_tool_calls: Vec::new(),
             max_call_repeats: 2,
             call_repeat_chain_len: 3,
@@ -170,6 +172,10 @@ impl SseHandler {
         message
     }
 
+    pub fn thinking_block(&mut self, block: ThinkingBlock) {
+        self.thinking.push(block);
+    }
+
     pub fn abort(&self) -> AbortSignal {
         self.abort_signal.clone()
     }
@@ -179,11 +185,14 @@ impl SseHandler {
         &self.last_tool_calls
     }
 
-    pub fn take(self) -> (String, Vec<ToolCall>) {
+    pub fn take(self) -> (String, Vec<ToolCall>, Vec<ThinkingBlock>) {
         let Self {
-            buffer, tool_calls, ..
+            buffer,
+            tool_calls,
+            thinking,
+            ..
         } = self;
-        (buffer, tool_calls)
+        (buffer, tool_calls, thinking)
     }
 }
 
@@ -232,8 +241,8 @@ where
         .map(|value| value.to_string());
     let is_event_stream = content_type
         .as_deref()
-        .map(|ct| ct.starts_with("text/event-stream"))
+        .map(|ct| ct.is_empty() || ct.starts_with("text/event-stream"))
-        .unwrap_or(false);
+        .unwrap_or(true);
     if !is_event_stream {
         let header_value = content_type.unwrap_or_default();
         let text = res.text().await?;

src/client/vertexai.rs

@@ -119,7 +119,11 @@ fn prepare_chat_completions(
             format!("{base_url}/google/models/{model_name}:{func}")
         }
         ModelCategory::Claude => {
-            format!("{base_url}/anthropic/models/{model_name}:streamRawPredict")
+            let func = match data.stream {
+                true => "streamRawPredict",
+                false => "rawPredict",
+            };
+            format!("{base_url}/anthropic/models/{model_name}:{func}")
         }
         ModelCategory::Mistral => {
             let func = match data.stream {
@@ -318,7 +322,11 @@ fn gemini_extract_chat_completions_text(data: &Value) -> Result<ChatCompletionsO
             bail!("Invalid response data: {data}");
         }
     }
-    let output = ChatCompletionsOutput { text, tool_calls };
+    let output = ChatCompletionsOutput {
+        text,
+        tool_calls,
+        ..Default::default()
+    };
     Ok(output)
 }
 
@@ -330,6 +338,7 @@ pub fn gemini_build_chat_completions_body(
         mut messages,
         temperature,
         top_p,
+        reasoning_effort,
         functions,
         stream: _,
     } = data;
@@ -367,8 +376,15 @@ pub fn gemini_build_chat_completions_body(
                             .collect();
                         vec![json!({ "role": role, "parts": parts })]
                     },
-                    MessageContent::ToolCalls(MessageContentToolCalls { tool_results, .. }) => {
+                    MessageContent::ToolCalls(MessageContentToolCalls { tool_results, text, .. }) => {
-                        let model_parts: Vec<Value> = tool_results.iter().map(|tool_result| {
+                        let mut model_parts: Vec<Value> = vec![];
+                        if !text.is_empty() {
+                            model_parts.push(json!({ "text": text }));
+                        }
+                        for tool_result in tool_results.iter() {
+                            if let Some(round_text) = &tool_result.text {
+                                model_parts.push(json!({ "text": round_text }));
+                            }
                             let mut part = json!({
                                 "functionCall": {
                                     "name": tool_result.call.name,
@@ -378,8 +394,8 @@ pub fn gemini_build_chat_completions_body(
                             if let Some(sig) = &tool_result.call.thought_signature {
                                 part["thoughtSignature"] = json!(sig);
                             }
-                            part
+                            model_parts.push(part);
-                        }).collect();
+                        }
                         let function_parts: Vec<Value> = tool_results.into_iter().map(|tool_result| {
                             json!({
                                 "functionResponse": {
@@ -422,6 +438,9 @@ pub fn gemini_build_chat_completions_body(
     if let Some(v) = top_p {
         body["generationConfig"]["topP"] = v.into();
     }
+    if let Some(v) = reasoning_effort {
+        body["generationConfig"]["thinking_config"] = json!({"thinking_level": v});
+    }
 
     if let Some(functions) = functions {
         // Gemini doesn't support functions with parameters that have empty properties, so we need to patch it.
@@ -479,7 +498,7 @@ pub async fn prepare_gcloud_access_token(
         let expires_at = Utc::now()
             + Duration::try_seconds(expires_in)
                 .ok_or_else(|| anyhow!("Failed to parse expires_in of access_token"))?;
-        set_access_token(client_name, token, expires_at.timestamp())
+        set_access_token(client_name, token, expires_at.timestamp(), None)
     }
     Ok(())
 }

src/config/agent.rs

@@ -2,6 +2,7 @@ use super::*;
 
 use crate::{
     client::Model,
+    config::memory,
     function::{Functions, run_llm_function},
 };
 
@@ -19,7 +20,7 @@ use fancy_regex::Captures;
 use inquire::{Text, validator::Validation};
 use rust_embed::Embed;
 use serde::{Deserialize, Serialize};
-use std::{ffi::OsStr, path::Path};
+use std::{env, ffi::OsStr, path::Path};
 
 const DEFAULT_AGENT_NAME: &str = "rag";
 
@@ -42,6 +43,8 @@ pub struct Agent {
     graph_rags: HashMap<String, Arc<Rag>>,
     model: Model,
     vault: GlobalVault,
+    is_graph: bool,
+    enabled_tools: Option<Vec<String>>,
 }
 
 impl Agent {
@@ -214,6 +217,20 @@ impl Agent {
             functions.append_skill_functions();
         }
 
+        if app.function_calling_support
+            && !matches!(agent_config.memory, Some(false))
+            && !matches!(app.memory, Some(false))
+        {
+            let memory_exists = paths::global_memory_index_file().exists()
+                || env::current_dir()
+                    .ok()
+                    .and_then(|cwd| memory::discover_workspace_memory(&cwd))
+                    .is_some();
+            if memory_exists {
+                functions.append_memory_functions();
+            }
+        }
+
         agent_config.replace_tools_placeholder(&functions);
 
         Ok(Self {
@@ -228,6 +245,8 @@ impl Agent {
             graph_rags,
             model,
             vault: app_state.vault.clone(),
+            is_graph: graph_for_rag.is_some(),
+            enabled_tools: None,
         })
     }
 
@@ -324,6 +343,10 @@ impl Agent {
         &self.name
     }
 
+    pub fn is_graph(&self) -> bool {
+        self.is_graph
+    }
+
     pub fn functions(&self) -> &Functions {
         &self.functions
     }
@@ -344,6 +367,10 @@ impl Agent {
         &self.config.mcp_servers
     }
 
+    pub fn spawnable_agents(&self) -> Option<&[String]> {
+        self.config.spawnable_agents.as_deref()
+    }
+
     pub fn skills_enabled(&self) -> Option<bool> {
         self.config.skills_enabled
     }
@@ -352,6 +379,10 @@ impl Agent {
         self.config.enabled_skills.as_deref()
     }
 
+    pub fn memory(&self) -> Option<bool> {
+        self.config.memory
+    }
+
     pub fn set_skills_enabled(&mut self, value: Option<bool>) {
         self.config.skills_enabled = value;
     }
@@ -464,6 +495,14 @@ impl Agent {
         self.config.continuation_prompt.clone()
     }
 
+    pub fn inject_skill_instructions(&self) -> bool {
+        self.config.inject_skill_instructions
+    }
+
+    pub fn skill_instructions_value(&self) -> Option<String> {
+        self.config.skill_instructions.clone()
+    }
+
     pub fn can_spawn_agents(&self) -> bool {
         self.config.can_spawn_agents
     }
@@ -492,6 +531,14 @@ impl Agent {
         self.config.compression_threshold
     }
 
+    pub fn max_tool_result_chars(&self) -> Option<usize> {
+        self.config.max_tool_result_chars
+    }
+
+    pub fn compression_keep_last(&self) -> Option<usize> {
+        self.config.compression_keep_last
+    }
+
     pub fn is_dynamic_instructions(&self) -> bool {
         self.config.dynamic_instructions
     }
@@ -548,8 +595,12 @@ impl RoleLike for Agent {
         self.config.top_p
     }
 
+    fn reasoning_effort(&self) -> Option<String> {
+        self.config.reasoning_effort.clone()
+    }
+
     fn enabled_tools(&self) -> Option<Vec<String>> {
-        None
+        self.enabled_tools.clone()
     }
 
     fn enabled_mcp_servers(&self) -> Option<Vec<String>> {
@@ -569,19 +620,18 @@ impl RoleLike for Agent {
         self.config.top_p = value;
     }
 
+    fn set_reasoning_effort(&mut self, value: Option<String>) {
+        self.config.reasoning_effort = value;
+    }
+
     fn set_enabled_tools(&mut self, value: Option<Vec<String>>) {
-        match value {
+        self.enabled_tools = value.map(|tools| {
-            Some(tools) => {
+            tools
-                self.config.global_tools = tools
                 .into_iter()
                 .map(|v| v.trim().to_string())
                 .filter(|v| !v.is_empty())
-                    .collect::<Vec<_>>();
+                .collect::<Vec<_>>()
-            }
+        });
-            None => {
-                self.config.global_tools.clear();
-            }
-        }
     }
 
     fn set_enabled_mcp_servers(&mut self, value: Option<Vec<String>>) {
@@ -610,11 +660,15 @@ pub struct AgentConfig {
     #[serde(skip_serializing_if = "Option::is_none")]
     pub top_p: Option<f64>,
     #[serde(skip_serializing_if = "Option::is_none")]
+    pub reasoning_effort: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
     pub agent_session: Option<String>,
     #[serde(default)]
     pub auto_continue: bool,
     #[serde(default)]
     pub can_spawn_agents: bool,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub spawnable_agents: Option<Vec<String>>,
     #[serde(default = "default_max_concurrent_agents")]
     pub max_concurrent_agents: usize,
     #[serde(default = "default_max_agent_depth")]
@@ -625,8 +679,18 @@ pub struct AgentConfig {
     pub inject_todo_instructions: bool,
     #[serde(default = "default_true")]
     pub inject_spawn_instructions: bool,
+    #[serde(default = "default_true")]
+    pub inject_skill_instructions: bool,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub skill_instructions: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub memory: Option<bool>,
     #[serde(skip_serializing_if = "Option::is_none")]
     pub compression_threshold: Option<usize>,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub max_tool_result_chars: Option<usize>,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub compression_keep_last: Option<usize>,
     #[serde(default)]
     pub description: String,
     #[serde(default)]
@@ -699,11 +763,14 @@ impl AgentConfig {
             model_id: graph.model.clone(),
             temperature: graph.temperature,
             top_p: graph.top_p,
+            reasoning_effort: graph.reasoning_effort.clone(),
             description: graph.description.clone(),
             global_tools: graph.global_tools.clone(),
             mcp_servers: graph.mcp_servers.clone(),
             skills_enabled: graph.skills_enabled,
             enabled_skills: graph.enabled_skills.clone(),
+            inject_skill_instructions: graph.inject_skill_instructions.unwrap_or(true),
+            skill_instructions: graph.skill_instructions.clone(),
             conversation_starters: graph.conversation_starters.clone(),
             variables: graph.variables.clone(),
             can_spawn_agents: graph.has_agent_node(),
@@ -731,6 +798,9 @@ impl AgentConfig {
         if let Some(v) = read_env_value::<f64>(&with_prefix("top_p")) {
             self.top_p = v;
         }
+        if let Some(v) = read_env_value::<String>(&with_prefix("reasoning_effort")) {
+            self.reasoning_effort = v;
+        }
         if let Ok(v) = env::var(with_prefix("global_tools"))
             && let Ok(v) = serde_json::from_str(&v)
         {
@@ -741,6 +811,11 @@ impl AgentConfig {
         {
             self.mcp_servers = v;
         }
+        if let Ok(v) = env::var(with_prefix("spawnable_agents"))
+            && let Ok(v) = serde_json::from_str(&v)
+        {
+            self.spawnable_agents = Some(v);
+        }
         if let Some(v) = read_env_value::<String>(&with_prefix("agent_session")) {
             self.agent_session = v;
         }
@@ -886,6 +961,9 @@ async fn init_graph_rags(
                 reranker_model: rag_node.reranker_model.clone(),
                 top_k: rag_node.top_k,
                 batch_size: rag_node.batch_size,
+                extractor_model: rag_node.extractor_model.clone(),
+                extractor_prompt: rag_node.extractor_prompt.clone(),
+                graph_hops: rag_node.graph_hops,
             };
             let fully_specified = config.embedding_model.is_some()
                 && config.chunk_size.is_some()
@@ -961,6 +1039,36 @@ pub fn list_agents() -> Vec<String> {
     agents
 }
 
+pub fn list_agents_with_descriptions() -> Vec<(String, String)> {
+    list_agents()
+        .into_iter()
+        .map(|name| {
+            let description = load_agent_description(&name);
+            (name, description)
+        })
+        .collect()
+}
+
+#[derive(Deserialize)]
+struct AgentMetadataStub {
+    #[serde(default)]
+    description: String,
+}
+
+fn load_agent_description(name: &str) -> String {
+    if let Ok(config) = AgentConfig::load(&paths::agent_config_file(name)) {
+        return config.description;
+    }
+
+    if let Ok(contents) = read_to_string(paths::agent_graph_file(name))
+        && let Ok(meta) = serde_yaml::from_str::<AgentMetadataStub>(&contents)
+    {
+        return meta.description;
+    }
+
+    String::new()
+}
+
 pub fn complete_agent_variables(agent_name: &str) -> Vec<(String, Option<String>)> {
     let config_path = paths::agent_config_file(agent_name);
     if !config_path.exists() {
@@ -1150,4 +1258,44 @@ variables:
         assert_eq!(config.max_agent_depth, default_max_agent_depth());
         assert_eq!(config.escalation_timeout, default_escalation_timeout());
     }
+
+    #[test]
+    fn agent_metadata_stub_extracts_description_from_graph_yaml() {
+        let yaml = r#"
+name: librarian
+description: External-reference research agent.
+version: "1.0"
+start: triage
+nodes: {}
+"#;
+
+        let meta: AgentMetadataStub = serde_yaml::from_str(yaml).unwrap();
+
+        assert_eq!(meta.description, "External-reference research agent.");
+    }
+
+    #[test]
+    fn agent_metadata_stub_extracts_multiline_description() {
+        let yaml = r#"
+name: coder
+description: |
+  Implementation agent. Plans, implements, and runs build + tests in a
+  bounded fix-loop until verified.
+version: "1.0"
+"#;
+
+        let meta: AgentMetadataStub = serde_yaml::from_str(yaml).unwrap();
+
+        assert!(meta.description.starts_with("Implementation agent."));
+        assert!(meta.description.contains("bounded fix-loop"));
+    }
+
+    #[test]
+    fn agent_metadata_stub_defaults_when_description_missing() {
+        let yaml = "name: nameless\nversion: \"1.0\"\n";
+
+        let meta: AgentMetadataStub = serde_yaml::from_str(yaml).unwrap();
+
+        assert_eq!(meta.description, "");
+    }
 }

src/config/app_config.rs

@@ -1,4 +1,4 @@
-use crate::client::{ClientConfig, list_models};
+use crate::client::{ClientConfig, Model, ModelType, list_models};
 use crate::render::{MarkdownRender, RenderOptions};
 use crate::utils::{IS_STDOUT_TERMINAL, NO_COLOR, decode_bin, get_env_name};
 
@@ -21,6 +21,7 @@ pub struct AppConfig {
     pub model_id: String,
     pub temperature: Option<f64>,
     pub top_p: Option<f64>,
+    pub reasoning_effort: Option<String>,
 
     pub dry_run: bool,
     pub stream: bool,
@@ -52,6 +53,8 @@ pub struct AppConfig {
     pub max_auto_continues: usize,
     pub inject_todo_instructions: bool,
     pub continuation_prompt: Option<String>,
+    pub inject_skill_instructions: bool,
+    pub skill_instructions: Option<String>,
 
     pub repl_prelude: Option<String>,
     pub cmd_prelude: Option<String>,
@@ -59,8 +62,17 @@ pub struct AppConfig {
 
     pub save_session: Option<bool>,
     pub compression_threshold: usize,
+    pub compression_keep_last: usize,
     pub summarization_prompt: Option<String>,
     pub summary_context_prompt: Option<String>,
+    pub max_tool_result_chars: Option<usize>,
+
+    pub memory: Option<bool>,
+    pub memory_cap_with_tools: Option<usize>,
+    pub memory_cap_without_tools: Option<usize>,
+
+    pub workspace_instructions: Option<bool>,
+    pub workspace_instructions_files: Option<Vec<String>>,
 
     pub rag_embedding_model: Option<String>,
     pub rag_reranker_model: Option<String>,
@@ -68,17 +80,22 @@ pub struct AppConfig {
     pub rag_chunk_size: Option<usize>,
     pub rag_chunk_overlap: Option<usize>,
     pub rag_template: Option<String>,
+    pub rag_extractor_model: Option<String>,
+    pub rag_extractor_prompt: Option<String>,
+    pub rag_graph_hops: usize,
 
     #[serde(default)]
     pub document_loaders: HashMap<String, String>,
 
     pub highlight: bool,
+    pub raw_markdown: bool,
     pub theme: Option<String>,
     pub left_prompt: Option<String>,
     pub right_prompt: Option<String>,
 
     pub user_agent: Option<String>,
     pub save_shell_history: bool,
+    pub no_workspace_mcp: bool,
     pub sync_models_url: Option<String>,
 
     pub clients: Vec<ClientConfig>,
@@ -90,6 +107,7 @@ impl Default for AppConfig {
             model_id: Default::default(),
             temperature: None,
             top_p: None,
+            reasoning_effort: None,
 
             dry_run: false,
             stream: true,
@@ -118,6 +136,8 @@ impl Default for AppConfig {
             max_auto_continues: 10,
             inject_todo_instructions: true,
             continuation_prompt: None,
+            inject_skill_instructions: true,
+            skill_instructions: None,
 
             repl_prelude: None,
             cmd_prelude: None,
@@ -125,8 +145,17 @@ impl Default for AppConfig {
 
             save_session: None,
             compression_threshold: 4000,
+            compression_keep_last: 0,
             summarization_prompt: None,
             summary_context_prompt: None,
+            max_tool_result_chars: None,
+
+            memory: None,
+            memory_cap_with_tools: None,
+            memory_cap_without_tools: None,
+
+            workspace_instructions: None,
+            workspace_instructions_files: None,
 
             rag_embedding_model: None,
             rag_reranker_model: None,
@@ -134,16 +163,21 @@ impl Default for AppConfig {
             rag_chunk_size: None,
             rag_chunk_overlap: None,
             rag_template: None,
+            rag_extractor_model: None,
+            rag_extractor_prompt: None,
+            rag_graph_hops: 1,
 
             document_loaders: Default::default(),
 
             highlight: true,
+            raw_markdown: false,
             theme: None,
             left_prompt: None,
             right_prompt: None,
 
             user_agent: None,
             save_shell_history: true,
+            no_workspace_mcp: false,
             sync_models_url: None,
 
             clients: vec![],
@@ -157,6 +191,7 @@ impl AppConfig {
             model_id: config.model_id,
             temperature: config.temperature,
             top_p: config.top_p,
+            reasoning_effort: None,
 
             dry_run: config.dry_run,
             stream: config.stream,
@@ -185,6 +220,8 @@ impl AppConfig {
             max_auto_continues: config.max_auto_continues,
             inject_todo_instructions: config.inject_todo_instructions,
             continuation_prompt: config.continuation_prompt,
+            inject_skill_instructions: config.inject_skill_instructions,
+            skill_instructions: config.skill_instructions,
 
             repl_prelude: config.repl_prelude,
             cmd_prelude: config.cmd_prelude,
@@ -192,8 +229,17 @@ impl AppConfig {
 
             save_session: config.save_session,
             compression_threshold: config.compression_threshold,
+            compression_keep_last: config.compression_keep_last,
             summarization_prompt: config.summarization_prompt,
             summary_context_prompt: config.summary_context_prompt,
+            max_tool_result_chars: config.max_tool_result_chars,
+
+            memory: config.memory,
+            memory_cap_with_tools: config.memory_cap_with_tools,
+            memory_cap_without_tools: config.memory_cap_without_tools,
+
+            workspace_instructions: config.workspace_instructions,
+            workspace_instructions_files: config.workspace_instructions_files,
 
             rag_embedding_model: config.rag_embedding_model,
             rag_reranker_model: config.rag_reranker_model,
@@ -201,16 +247,21 @@ impl AppConfig {
             rag_chunk_size: config.rag_chunk_size,
             rag_chunk_overlap: config.rag_chunk_overlap,
             rag_template: config.rag_template,
+            rag_extractor_model: config.rag_extractor_model,
+            rag_extractor_prompt: config.rag_extractor_prompt,
+            rag_graph_hops: config.rag_graph_hops,
 
             document_loaders: config.document_loaders,
 
             highlight: config.highlight,
+            raw_markdown: config.raw_markdown,
             theme: config.theme,
             left_prompt: config.left_prompt,
             right_prompt: config.right_prompt,
 
             user_agent: config.user_agent,
             save_shell_history: config.save_shell_history,
+            no_workspace_mcp: false,
             sync_models_url: config.sync_models_url,
 
             clients: config.clients,
@@ -223,6 +274,7 @@ impl AppConfig {
         app_config.setup_document_loaders();
         app_config.setup_user_agent();
         app_config.resolve_model()?;
+        app_config.validate_reasoning_effort()?;
         Ok(app_config)
     }
 
@@ -243,6 +295,31 @@ impl AppConfig {
         Ok(())
     }
 
+    fn validate_reasoning_effort(&self) -> Result<()> {
+        let Some(ref effort) = self.reasoning_effort else {
+            return Ok(());
+        };
+        let model = Model::retrieve_model(self, &self.model_id, ModelType::Chat)?;
+        let levels = model.reasoning_levels();
+
+        if levels.is_empty() {
+            bail!(
+                "reasoning_effort '{}' is configured but the model does not support reasoning effort",
+                effort
+            );
+        }
+
+        if !levels.iter().any(|l| l == effort) {
+            bail!(
+                "reasoning_effort '{}' is not valid for the model. Supported levels: {}",
+                effort,
+                levels.join(", ")
+            );
+        }
+
+        Ok(())
+    }
+
     pub fn resolve_model(&mut self) -> Result<()> {
         if self.model_id.is_empty() {
             let models = list_models(self, crate::client::ModelType::Chat);
@@ -256,26 +333,43 @@ impl AppConfig {
 
     pub fn vault_password_file(&self) -> PathBuf {
         match &self.vault_password_file {
-            Some(path) => match path.exists() {
+            Some(path) => {
-                true => path.clone(),
+                if path.exists() {
-                false => gman::config::Config::local_provider_password_file(),
+                    return path.clone();
-            },
+                }
+
+                if let Some(translated) = paths::translate_sandboxed_home_dir(path)
+                    && translated.exists()
+                {
+                    info!(
+                        "vault_password_file '{}' not found; resolved to sandboxed path '{}'",
+                        path.display(),
+                        translated.display()
+                    );
+
+                    return translated;
+                }
+
+                gman::config::Config::local_provider_password_file()
+            }
             None => gman::config::Config::local_provider_password_file(),
         }
     }
 
     pub fn editor(&self) -> Result<String> {
         super::EDITOR.get_or_init(move || {
-            let editor = self.editor.clone()
+            if let Some(editor) = self.editor.clone()
                 .or_else(|| env::var("VISUAL").ok().or_else(|| env::var("EDITOR").ok()))
-                .unwrap_or_else(|| {
+                && which::which(&editor).is_ok()
-                    if cfg!(windows) {
+            {
+                return Some(editor);
+            }
+            let default = if cfg!(windows) {
                 "notepad".to_string()
             } else {
                 "nano".to_string()
-                    }
+            };
-                });
+            which::which(&default).ok().map(|_| default)
-            which::which(&editor).ok().map(|_| editor)
         })
             .clone()
             .ok_or_else(|| anyhow!("Editor not found. Please add the `editor` configuration or set the $EDITOR or $VISUAL environment variable."))
@@ -295,7 +389,7 @@ impl AppConfig {
         let theme = if self.highlight {
             let theme_mode = if self.light_theme() { "light" } else { "dark" };
             let theme_filename = format!("{theme_mode}.tmTheme");
-            let theme_path = paths::local_path(&theme_filename);
+            let theme_path = paths::local_dir(&theme_filename);
             if theme_path.exists() {
                 let theme = ThemeSet::get_theme(&theme_path)
                     .with_context(|| format!("Invalid theme at '{}'", theme_path.display()))?;
@@ -320,14 +414,26 @@ impl AppConfig {
             env::var("COLORTERM").as_ref().map(|v| v.as_str()),
             Ok("truecolor")
         );
-        Ok(RenderOptions::new(theme, wrap, self.wrap_code, truecolor))
+        Ok(RenderOptions::new(
+            theme,
+            wrap,
+            self.wrap_code,
+            self.raw_markdown,
+            truecolor,
+        ))
     }
 
     pub fn print_markdown(&self, text: &str) -> Result<()> {
         if *IS_STDOUT_TERMINAL {
             let render_options = self.render_options()?;
             let mut markdown_render = MarkdownRender::init(render_options)?;
-            println!("{}", markdown_render.render(text));
+            let body = markdown_render.render(text);
+            let tail = markdown_render.finalize();
+            if tail.is_empty() {
+                println!("{body}");
+            } else {
+                println!("{body}\n{tail}");
+            }
         } else {
             println!("{text}");
         }
@@ -379,6 +485,9 @@ impl AppConfig {
         if let Some(v) = super::read_env_value::<f64>(&get_env_name("top_p")) {
             self.top_p = v;
         }
+        if let Some(v) = super::read_env_value::<String>(&get_env_name("reasoning_effort")) {
+            self.reasoning_effort = v;
+        }
 
         if let Some(Some(v)) = super::read_env_bool(&get_env_name("dry_run")) {
             self.dry_run = v;
@@ -479,6 +588,15 @@ impl AppConfig {
         if let Some(v) = super::read_env_value::<String>(&get_env_name("rag_template")) {
             self.rag_template = v;
         }
+        if let Some(v) = super::read_env_value::<String>(&get_env_name("rag_extractor_model")) {
+            self.rag_extractor_model = v;
+        }
+        if let Some(v) = super::read_env_value::<String>(&get_env_name("rag_extractor_prompt")) {
+            self.rag_extractor_prompt = v;
+        }
+        if let Some(v) = super::read_env_value::<usize>(&get_env_name("rag_graph_hops")) {
+            self.rag_graph_hops = v.unwrap_or(1);
+        }
 
         if let Ok(v) = env::var(get_env_name("document_loaders"))
             && let Ok(v) = serde_json::from_str(&v)
@@ -492,6 +610,9 @@ impl AppConfig {
         if *NO_COLOR {
             self.highlight = false;
         }
+        if let Some(Some(v)) = super::read_env_bool(&get_env_name("raw_markdown")) {
+            self.raw_markdown = v;
+        }
         if self.highlight && self.theme.is_none() {
             if let Some(v) = super::read_env_value::<String>(&get_env_name("theme")) {
                 self.theme = v;

src/config/input.rs

@@ -253,6 +253,10 @@ impl Input {
         patch_messages(&mut messages, model);
         model.guard_max_input_tokens(&messages)?;
         let (temperature, top_p) = (self.role().temperature(), self.role().top_p());
+        let reasoning_effort = self
+            .role()
+            .reasoning_effort()
+            .or_else(|| model.default_reasoning_effort().map(|s| s.to_string()));
         let functions = if model.supports_function_calling() {
             let fns = self.functions.clone();
             if let Some(vec) = &fns {
@@ -268,6 +272,7 @@ impl Input {
             messages,
             temperature,
             top_p,
+            reasoning_effort,
             functions,
             stream,
         })

src/config/instructions.rs

@@ -0,0 +1,211 @@
+use std::fs;
+use std::path::{Path, PathBuf};
+
+use log::warn;
+
+pub const WORKSPACE_INSTRUCTIONS_FILE_NAME: &str = "COYOTE.md";
+pub const DEFAULT_WORKSPACE_INSTRUCTIONS_FILES: [&str; 4] = [
+    WORKSPACE_INSTRUCTIONS_FILE_NAME,
+    "AGENTS.md",
+    "CLAUDE.md",
+    "GEMINI.md",
+];
+const INSTRUCTIONS_SIZE_WARN_THRESHOLD: usize = 24_000;
+
+#[derive(Debug, Clone)]
+pub struct WorkspaceInstructions {
+    pub path: PathBuf,
+    pub content: String,
+}
+
+pub fn default_workspace_instructions_files() -> Vec<String> {
+    DEFAULT_WORKSPACE_INSTRUCTIONS_FILES
+        .iter()
+        .map(|s| s.to_string())
+        .collect()
+}
+
+pub fn discover_workspace_instructions(
+    start: &Path,
+    file_names: &[String],
+) -> Option<WorkspaceInstructions> {
+    for dir in start.ancestors() {
+        for name in file_names {
+            let candidate = dir.join(name);
+            if !candidate.is_file() {
+                continue;
+            }
+            match fs::read_to_string(&candidate) {
+                Ok(content) if !content.trim().is_empty() => {
+                    return Some(WorkspaceInstructions {
+                        path: candidate,
+                        content,
+                    });
+                }
+                Ok(_) => {}
+                Err(e) => warn!(
+                    "failed to read workspace instructions at {}: {e}",
+                    candidate.display()
+                ),
+            }
+        }
+    }
+    None
+}
+
+pub fn build_instructions_section(instructions: &WorkspaceInstructions) -> String {
+    let char_count = instructions.content.chars().count();
+    if char_count > INSTRUCTIONS_SIZE_WARN_THRESHOLD {
+        warn!(
+            "workspace instructions at {} are large ({char_count} chars); \
+             consider moving detail into workspace memory drill files",
+            instructions.path.display()
+        );
+    }
+
+    format!(
+        "<workspace_instructions source=\"{}\">\n{}\n</workspace_instructions>",
+        instructions.path.display(),
+        instructions.content.trim_end()
+    )
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::{env, time};
+    use time::SystemTime;
+
+    fn temp_root(label: &str) -> PathBuf {
+        let unique = SystemTime::now()
+            .duration_since(time::UNIX_EPOCH)
+            .unwrap()
+            .as_nanos();
+        let root = env::temp_dir().join(format!("coyote-instructions-{label}-{unique}"));
+        fs::create_dir_all(&root).unwrap();
+        root
+    }
+
+    fn defaults() -> Vec<String> {
+        default_workspace_instructions_files()
+    }
+
+    #[test]
+    fn discovery_returns_none_when_no_file_exists() {
+        let root = temp_root("none");
+
+        assert!(discover_workspace_instructions(&root, &defaults()).is_none());
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn discovery_finds_coyote_md() {
+        let root = temp_root("coyote");
+        fs::write(root.join("COYOTE.md"), "coyote instructions").unwrap();
+
+        let found = discover_workspace_instructions(&root, &defaults()).unwrap();
+        assert_eq!(found.path, root.join("COYOTE.md"));
+        assert_eq!(found.content, "coyote instructions");
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn discovery_falls_back_through_chain_in_order() {
+        let root = temp_root("fallback");
+        fs::write(root.join("GEMINI.md"), "gemini instructions").unwrap();
+
+        let found = discover_workspace_instructions(&root, &defaults()).unwrap();
+        assert_eq!(found.path, root.join("GEMINI.md"));
+
+        fs::write(root.join("CLAUDE.md"), "claude instructions").unwrap();
+        let found = discover_workspace_instructions(&root, &defaults()).unwrap();
+        assert_eq!(found.path, root.join("CLAUDE.md"));
+
+        fs::write(root.join("AGENTS.md"), "agents instructions").unwrap();
+        let found = discover_workspace_instructions(&root, &defaults()).unwrap();
+        assert_eq!(found.path, root.join("AGENTS.md"));
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn discovery_prefers_coyote_md_over_fallbacks() {
+        let root = temp_root("precedence");
+        fs::write(root.join("COYOTE.md"), "coyote").unwrap();
+        fs::write(root.join("AGENTS.md"), "agents").unwrap();
+        fs::write(root.join("CLAUDE.md"), "claude").unwrap();
+
+        let found = discover_workspace_instructions(&root, &defaults()).unwrap();
+        assert_eq!(found.path, root.join("COYOTE.md"));
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn discovery_walks_up_from_nested_dir() {
+        let root = temp_root("walk_up");
+        fs::write(root.join("AGENTS.md"), "root instructions").unwrap();
+        let nested = root.join("src").join("deep");
+        fs::create_dir_all(&nested).unwrap();
+
+        let found = discover_workspace_instructions(&nested, &defaults()).unwrap();
+        assert_eq!(found.path, root.join("AGENTS.md"));
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn discovery_prefers_closer_file_over_higher_priority_name_above() {
+        let root = temp_root("depth_first");
+        fs::write(root.join("COYOTE.md"), "root coyote").unwrap();
+        let nested = root.join("packages").join("app");
+        fs::create_dir_all(&nested).unwrap();
+        fs::write(nested.join("CLAUDE.md"), "nested claude").unwrap();
+
+        let found = discover_workspace_instructions(&nested, &defaults()).unwrap();
+        assert_eq!(found.path, nested.join("CLAUDE.md"));
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn discovery_skips_empty_files() {
+        let root = temp_root("empty");
+        fs::write(root.join("COYOTE.md"), "  \n").unwrap();
+        fs::write(root.join("AGENTS.md"), "real content").unwrap();
+
+        let found = discover_workspace_instructions(&root, &defaults()).unwrap();
+        assert_eq!(found.path, root.join("AGENTS.md"));
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn discovery_honors_custom_file_chain() {
+        let root = temp_root("custom");
+        fs::write(root.join("CLAUDE.md"), "claude").unwrap();
+
+        let only_agents = vec!["AGENTS.md".to_string()];
+        assert!(discover_workspace_instructions(&root, &only_agents).is_none());
+
+        let empty: Vec<String> = vec![];
+        assert!(discover_workspace_instructions(&root, &empty).is_none());
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn build_section_wraps_content_with_source_path() {
+        let instructions = WorkspaceInstructions {
+            path: PathBuf::from("/ws/COYOTE.md"),
+            content: "Do the thing.\n".into(),
+        };
+
+        let section = build_instructions_section(&instructions);
+        assert!(section.starts_with("<workspace_instructions source=\"/ws/COYOTE.md\">"));
+        assert!(section.contains("Do the thing."));
+        assert!(section.ends_with("</workspace_instructions>"));
+    }
+}

src/config/mcp_factory.rs

@@ -1,4 +1,7 @@
-use crate::mcp::{ConnectedServer, JsonField, McpServer, McpTransportType, spawn_mcp_server};
+use crate::mcp::{
+    ConnectedServer, JsonField, McpServer, McpTransportType, is_auth_required_error, oauth,
+    spawn_mcp_server,
+};
 
 use anyhow::Result;
 use parking_lot::Mutex;
@@ -99,7 +102,23 @@ impl McpFactory {
             return Ok(existing);
         }
 
-        let handle = spawn_mcp_server(spec, log_path).await?;
+        let bearer_token = if spec.is_remote() {
+            oauth::load_valid_mcp_token(name)
+        } else {
+            None
+        };
+        let handle = spawn_mcp_server(spec, log_path, bearer_token)
+            .await
+            .map_err(|e| {
+                if is_auth_required_error(&e) {
+                    e.context(format!(
+                        "MCP server '{name}' requires OAuth authentication. \
+                         Run `coyote --auth-mcp {name}` or `.mcp auth {name}` in the REPL to authenticate."
+                    ))
+                } else {
+                    e
+                }
+            })?;
         self.insert_active(key, &handle);
         Ok(handle)
     }
@@ -125,6 +144,7 @@ mod tests {
             cwd: None,
             url: None,
             headers: None,
+            oauth: None,
         }
     }
 
@@ -141,6 +161,7 @@ mod tests {
             cwd: None,
             url: Some(url.to_string()),
             headers,
+            oauth: None,
         }
     }
 

src/config/memory.rs

@@ -0,0 +1,722 @@
+use std::fs;
+use std::path::{Path, PathBuf};
+
+use anyhow::{Context, Result};
+use log::warn;
+use serde::{Deserialize, Serialize};
+
+use crate::config::{
+    GIT_DIR_NAME, GITIGNORE_FILE_NAME, MEMORY_DIR_NAME, MEMORY_INDEX_FILE_NAME,
+    WORKSPACE_COYOTE_DIR_NAME, paths,
+};
+
+pub const DEFAULT_MEMORY_CAP_WITH_TOOLS: usize = 6_000;
+pub const DEFAULT_MEMORY_CAP_WITHOUT_TOOLS: usize = 12_000;
+
+#[derive(Debug, Clone)]
+pub struct WorkspaceMemory {
+    pub workspace_root: PathBuf,
+    pub dir: PathBuf,
+}
+
+pub fn discover_workspace_memory(start: &Path) -> Option<WorkspaceMemory> {
+    for dir in start.ancestors() {
+        let structured = dir.join(WORKSPACE_COYOTE_DIR_NAME).join(MEMORY_DIR_NAME);
+        if structured.join(MEMORY_INDEX_FILE_NAME).exists() {
+            return Some(WorkspaceMemory {
+                workspace_root: dir.to_path_buf(),
+                dir: structured,
+            });
+        }
+    }
+    None
+}
+
+pub fn find_git_root(start: &Path) -> Option<PathBuf> {
+    for dir in start.ancestors() {
+        if dir.join(GIT_DIR_NAME).exists() {
+            return Some(dir.to_path_buf());
+        }
+    }
+
+    None
+}
+
+pub fn bootstrap_workspace_memory(git_root: &Path) -> Result<PathBuf> {
+    let mem_dir = paths::workspace_memory_dir_for(git_root);
+    fs::create_dir_all(&mem_dir)
+        .with_context(|| format!("create memory dir {}", mem_dir.display()))?;
+
+    let index_path = mem_dir.join(MEMORY_INDEX_FILE_NAME);
+    if !index_path.exists() {
+        fs::write(&index_path, "# Workspace Memory Index\n\n")
+            .with_context(|| format!("write {}", index_path.display()))?;
+    }
+
+    let gitignore_appended = append_gitignore_entry(git_root)?;
+    let suffix = if gitignore_appended {
+        " (appended .coyote/memory/ to .gitignore)"
+    } else {
+        ""
+    };
+    warn!(
+        "auto-bootstrapped workspace memory at {}{}",
+        mem_dir.display(),
+        suffix
+    );
+
+    Ok(mem_dir)
+}
+
+pub fn append_gitignore_entry(git_root: &Path) -> Result<bool> {
+    let gitignore = git_root.join(GITIGNORE_FILE_NAME);
+    let entry = format!("{WORKSPACE_COYOTE_DIR_NAME}/{MEMORY_DIR_NAME}/");
+    let entry_no_slash = format!("{WORKSPACE_COYOTE_DIR_NAME}/{MEMORY_DIR_NAME}");
+
+    let existing = fs::read_to_string(&gitignore).unwrap_or_default();
+    let already_present = existing.lines().any(|line| {
+        let trimmed = line.trim();
+        trimmed == entry || trimmed == entry_no_slash
+    });
+
+    if already_present {
+        return Ok(false);
+    }
+
+    let new_content = if existing.is_empty() {
+        format!("{entry}\n")
+    } else if existing.ends_with('\n') {
+        format!("{existing}{entry}\n")
+    } else {
+        format!("{existing}\n{entry}\n")
+    };
+
+    fs::write(&gitignore, new_content).with_context(|| format!("write {}", gitignore.display()))?;
+
+    Ok(true)
+}
+
+#[derive(Debug, Clone, Deserialize, Serialize, Default)]
+pub struct MemoryFrontmatter {
+    #[serde(default)]
+    pub name: String,
+    #[serde(default)]
+    pub description: Option<String>,
+    #[serde(default, rename = "type")]
+    pub kind: Option<String>,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub created: Option<String>,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub updated: Option<String>,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub superseded_by: Option<String>,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub expires: Option<String>,
+}
+
+#[derive(Debug, Clone)]
+pub struct MemoryFile {
+    pub path: PathBuf,
+    pub frontmatter: MemoryFrontmatter,
+    pub body: String,
+}
+
+impl MemoryFile {
+    pub fn load(path: &Path) -> Result<Self> {
+        let raw = fs::read_to_string(path)
+            .with_context(|| format!("read memory file {}", path.display()))?;
+        let (frontmatter, body) = parse_frontmatter(&raw)
+            .with_context(|| format!("parse frontmatter in {}", path.display()))?;
+
+        Ok(Self {
+            path: path.to_path_buf(),
+            frontmatter,
+            body,
+        })
+    }
+
+    pub fn save(&self) -> Result<()> {
+        if let Some(parent) = self.path.parent() {
+            fs::create_dir_all(parent)?;
+        }
+
+        let frontmatter_yaml = serde_yaml::to_string(&self.frontmatter)?;
+        let content = format!("---\n{}---\n\n{}", frontmatter_yaml, self.body);
+
+        fs::write(&self.path, content)?;
+
+        Ok(())
+    }
+
+    pub fn char_len(&self) -> usize {
+        self.body.chars().count()
+    }
+}
+
+fn parse_frontmatter(raw: &str) -> Result<(MemoryFrontmatter, String)> {
+    let trimmed = raw.trim_start();
+    if !trimmed.starts_with("---") {
+        return Ok((MemoryFrontmatter::default(), raw.to_string()));
+    }
+
+    let after = &trimmed[3..];
+    let Some(end) = after.find("\n---") else {
+        return Ok((MemoryFrontmatter::default(), raw.to_string()));
+    };
+    let yaml = &after[..end];
+    let body = after[end + 4..].trim_start_matches('\n').to_string();
+    let frontmatter: MemoryFrontmatter =
+        serde_yaml::from_str(yaml.trim()).context("parse YAML frontmatter")?;
+
+    Ok((frontmatter, body))
+}
+
+#[derive(Debug, Clone)]
+pub struct MemoryStore {
+    pub global_dir: PathBuf,
+    pub workspace: Option<WorkspaceMemory>,
+}
+
+impl MemoryStore {
+    pub fn new(cwd: &Path) -> Self {
+        Self {
+            global_dir: paths::global_memory_dir(),
+            workspace: discover_workspace_memory(cwd),
+        }
+    }
+
+    pub fn load_global_index(&self) -> Result<Option<String>> {
+        let path = self.global_dir.join(MEMORY_INDEX_FILE_NAME);
+
+        if path.exists() {
+            Ok(Some(fs::read_to_string(path)?))
+        } else {
+            Ok(None)
+        }
+    }
+
+    pub fn load_workspace_index(&self) -> Result<Option<String>> {
+        match &self.workspace {
+            None => Ok(None),
+            Some(ws) => {
+                let index = ws.dir.join(MEMORY_INDEX_FILE_NAME);
+                if index.exists() {
+                    Ok(Some(fs::read_to_string(index)?))
+                } else {
+                    Ok(None)
+                }
+            }
+        }
+    }
+
+    pub fn list_files(&self) -> Result<Vec<MemoryFile>> {
+        let mut out = Vec::new();
+
+        if self.global_dir.exists() {
+            collect_md_files(&self.global_dir, &mut out)?;
+        }
+
+        if let Some(ws) = &self.workspace {
+            collect_md_files(&ws.dir, &mut out)?;
+        }
+
+        Ok(out)
+    }
+}
+
+pub fn build_memory_section(
+    store: &MemoryStore,
+    with_tools: bool,
+    cap: usize,
+) -> Result<Option<String>> {
+    let global_index = store.load_global_index()?;
+    let workspace_index = store.load_workspace_index()?;
+
+    if global_index.is_none() && workspace_index.is_none() {
+        return Ok(None);
+    }
+
+    let mut buf = String::from("<memory>\n");
+    let mut consumed = 0usize;
+
+    if let Some(s) = &global_index {
+        buf.push_str("<global_index>\n");
+        buf.push_str(s);
+        buf.push_str("\n</global_index>\n");
+        consumed += s.chars().count();
+    }
+
+    if let Some(s) = &workspace_index {
+        buf.push_str("<workspace_index>\n");
+        buf.push_str(s);
+        buf.push_str("\n</workspace_index>\n");
+        consumed += s.chars().count();
+    }
+
+    if consumed > cap {
+        warn!(
+            "memory indexes ({} chars) exceed cap ({} chars); injecting fully - \
+             consider raising memory_cap_* in config or shrinking MEMORY.md",
+            consumed, cap
+        );
+    }
+
+    if !with_tools {
+        let mut budget = cap.saturating_sub(consumed);
+        let mut files = store.list_files()?;
+        files.sort_by(|a, b| a.frontmatter.name.cmp(&b.frontmatter.name));
+        let mut omitted = 0usize;
+        for f in files {
+            let needed = f.body.chars().count() + 50;
+            if needed > budget {
+                omitted += 1;
+                continue;
+            }
+            buf.push_str(&format!("<file name=\"{}\">\n", f.frontmatter.name));
+            buf.push_str(&f.body);
+            buf.push_str("\n</file>\n");
+            budget = budget.saturating_sub(needed);
+        }
+
+        if omitted > 0 {
+            buf.push_str(&format!(
+                "<!-- {} memory file(s) omitted; enable function calling for full access -->\n",
+                omitted
+            ));
+        }
+    }
+
+    buf.push_str("</memory>");
+    Ok(Some(buf))
+}
+
+fn collect_md_files(dir: &Path, out: &mut Vec<MemoryFile>) -> Result<()> {
+    for entry in fs::read_dir(dir)? {
+        let entry = entry?;
+        let path = entry.path();
+        if path.extension().and_then(|e| e.to_str()) != Some("md") {
+            continue;
+        }
+
+        if path.file_name().and_then(|n| n.to_str()) == Some(MEMORY_INDEX_FILE_NAME) {
+            continue;
+        }
+
+        match MemoryFile::load(&path) {
+            Ok(f) => out.push(f),
+            Err(e) => warn!("skip malformed memory file {}: {}", path.display(), e),
+        }
+    }
+
+    Ok(())
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::{env, time};
+    use time::SystemTime;
+
+    fn temp_root(label: &str) -> PathBuf {
+        let unique = SystemTime::now()
+            .duration_since(time::UNIX_EPOCH)
+            .unwrap()
+            .as_nanos();
+        let root = env::temp_dir().join(format!("coyote-memory-{label}-{unique}"));
+        fs::create_dir_all(&root).unwrap();
+        root
+    }
+
+    #[test]
+    fn loads_global_and_workspace_indexes_from_test_dirs() {
+        let root = temp_root("phase1");
+        let workspace = root.join("workspace");
+        let workspace_memory_dir = workspace
+            .join(WORKSPACE_COYOTE_DIR_NAME)
+            .join(MEMORY_DIR_NAME);
+        fs::create_dir_all(&workspace_memory_dir).unwrap();
+        fs::write(
+            workspace_memory_dir.join(MEMORY_INDEX_FILE_NAME),
+            "workspace-content",
+        )
+        .unwrap();
+
+        let global = root.join("global");
+        fs::create_dir_all(&global).unwrap();
+        fs::write(global.join(MEMORY_INDEX_FILE_NAME), "global-content").unwrap();
+
+        let store = MemoryStore {
+            global_dir: global,
+            workspace: discover_workspace_memory(&workspace),
+        };
+
+        assert_eq!(
+            store.load_global_index().unwrap().as_deref(),
+            Some("global-content")
+        );
+        assert_eq!(
+            store.load_workspace_index().unwrap().as_deref(),
+            Some("workspace-content")
+        );
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn workspace_discovery_ignores_root_instructions_file() {
+        let root = temp_root("no_lite");
+        let workspace = root.join("ws");
+        fs::create_dir_all(&workspace).unwrap();
+        fs::write(workspace.join("COYOTE.md"), "instructions, not memory").unwrap();
+
+        assert!(discover_workspace_memory(&workspace).is_none());
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn build_memory_section_returns_none_when_no_memory_exists() {
+        let root = temp_root("none");
+        let workspace = root.join("ws");
+        fs::create_dir_all(&workspace).unwrap();
+
+        let store = MemoryStore {
+            global_dir: root.join("global"),
+            workspace: discover_workspace_memory(&workspace),
+        };
+
+        assert!(build_memory_section(&store, true, 6_000).unwrap().is_none());
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn build_memory_section_injects_only_indexes_with_tools_on() {
+        let root = temp_root("indexes_only");
+        let workspace = root.join("ws");
+        let structured = workspace
+            .join(WORKSPACE_COYOTE_DIR_NAME)
+            .join(MEMORY_DIR_NAME);
+        fs::create_dir_all(&structured).unwrap();
+        fs::write(
+            structured.join(MEMORY_INDEX_FILE_NAME),
+            "workspace-index-content",
+        )
+        .unwrap();
+        fs::write(
+            structured.join("foo.md"),
+            "---\nname: foo\n---\nfoo body that should not appear\n",
+        )
+        .unwrap();
+
+        let store = MemoryStore {
+            global_dir: root.join("global"),
+            workspace: discover_workspace_memory(&workspace),
+        };
+
+        let section = build_memory_section(&store, true, 6_000)
+            .unwrap()
+            .expect("memory section should exist");
+        assert!(section.contains("workspace-index-content"));
+        assert!(!section.contains("foo body that should not appear"));
+        assert!(section.starts_with("<memory>"));
+        assert!(section.ends_with("</memory>"));
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn build_memory_section_injects_drill_bodies_alphabetically_without_tools() {
+        let root = temp_root("drill_bodies");
+        let workspace = root.join("ws");
+        let structured = workspace
+            .join(WORKSPACE_COYOTE_DIR_NAME)
+            .join(MEMORY_DIR_NAME);
+        fs::create_dir_all(&structured).unwrap();
+        fs::write(structured.join(MEMORY_INDEX_FILE_NAME), "idx").unwrap();
+        fs::write(
+            structured.join("zebra.md"),
+            "---\nname: zebra\n---\nzebra body\n",
+        )
+        .unwrap();
+        fs::write(
+            structured.join("alpha.md"),
+            "---\nname: alpha\n---\nalpha body\n",
+        )
+        .unwrap();
+
+        let store = MemoryStore {
+            global_dir: root.join("global"),
+            workspace: discover_workspace_memory(&workspace),
+        };
+
+        let section = build_memory_section(&store, false, 6_000)
+            .unwrap()
+            .expect("memory section should exist");
+        let alpha_pos = section.find("alpha body").expect("alpha body missing");
+        let zebra_pos = section.find("zebra body").expect("zebra body missing");
+        assert!(alpha_pos < zebra_pos, "drill bodies must be alphabetical");
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn build_memory_section_omits_drill_bodies_when_cap_exceeded() {
+        let root = temp_root("cap");
+        let workspace = root.join("ws");
+        let structured = workspace
+            .join(WORKSPACE_COYOTE_DIR_NAME)
+            .join(MEMORY_DIR_NAME);
+        fs::create_dir_all(&structured).unwrap();
+        fs::write(structured.join(MEMORY_INDEX_FILE_NAME), "idx").unwrap();
+        let big_body = "x".repeat(200);
+        fs::write(
+            structured.join("big.md"),
+            format!("---\nname: big\n---\n{}\n", big_body),
+        )
+        .unwrap();
+
+        let store = MemoryStore {
+            global_dir: root.join("global"),
+            workspace: discover_workspace_memory(&workspace),
+        };
+
+        let section = build_memory_section(&store, false, 100)
+            .unwrap()
+            .expect("memory section should exist");
+        assert!(!section.contains(&big_body));
+        assert!(section.contains("memory file(s) omitted"));
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn parse_frontmatter_extracts_yaml() {
+        let raw = "---\nname: foo\ndescription: a thing\ntype: user\n---\nBody text\n";
+
+        let (fm, body) = parse_frontmatter(raw).unwrap();
+
+        assert_eq!(fm.name, "foo");
+        assert_eq!(fm.description.as_deref(), Some("a thing"));
+        assert_eq!(fm.kind.as_deref(), Some("user"));
+        assert_eq!(body, "Body text\n");
+    }
+
+    #[test]
+    fn parse_frontmatter_handles_missing_block() {
+        let raw = "# Just markdown, no frontmatter\nbody";
+
+        let (fm, body) = parse_frontmatter(raw).unwrap();
+
+        assert_eq!(fm.name, "");
+        assert!(fm.kind.is_none());
+        assert_eq!(body, raw);
+    }
+
+    #[test]
+    fn parse_frontmatter_handles_unterminated_block() {
+        let raw = "---\nname: oops\nno closing delimiter\n# rest of doc";
+
+        let (fm, body) = parse_frontmatter(raw).unwrap();
+
+        assert_eq!(fm.name, "");
+        assert_eq!(body, raw);
+    }
+
+    #[test]
+    fn memory_file_save_and_load_roundtrip() {
+        let root = temp_root("roundtrip");
+        let path = root.join("test.md");
+        let file = MemoryFile {
+            path: path.clone(),
+            frontmatter: MemoryFrontmatter {
+                name: "test".into(),
+                description: Some("a test".into()),
+                kind: Some("user".into()),
+                ..Default::default()
+            },
+            body: "Hello world\nmore text".into(),
+        };
+        file.save().unwrap();
+        let loaded = MemoryFile::load(&path).unwrap();
+        assert_eq!(loaded.frontmatter.name, "test");
+        assert_eq!(loaded.frontmatter.description.as_deref(), Some("a test"));
+        assert_eq!(loaded.frontmatter.kind.as_deref(), Some("user"));
+        assert_eq!(loaded.body, "Hello world\nmore text");
+
+        let raw = fs::read_to_string(&path).unwrap();
+        assert!(raw.contains("type: user"), "kind must serialize as 'type:'");
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn discover_walks_up_from_nested_dir() {
+        let root = temp_root("walk_up");
+        let workspace = root.join("ws");
+        let mem_dir = workspace
+            .join(WORKSPACE_COYOTE_DIR_NAME)
+            .join(MEMORY_DIR_NAME);
+        fs::create_dir_all(&mem_dir).unwrap();
+        fs::write(mem_dir.join(MEMORY_INDEX_FILE_NAME), "idx").unwrap();
+        let nested = workspace.join("src").join("deep").join("path");
+        fs::create_dir_all(&nested).unwrap();
+
+        let found = discover_workspace_memory(&nested).expect("workspace memory should be found");
+        assert_eq!(found.dir, mem_dir);
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn find_git_root_returns_dir_containing_git_dir() {
+        let root = temp_root("git_root");
+        let repo = root.join("repo");
+        fs::create_dir_all(repo.join(GIT_DIR_NAME)).unwrap();
+
+        assert_eq!(find_git_root(&repo), Some(repo.clone()));
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn find_git_root_walks_up_from_nested_dir() {
+        let root = temp_root("git_root_walk");
+        let repo = root.join("repo");
+        fs::create_dir_all(repo.join(GIT_DIR_NAME)).unwrap();
+        let nested = repo.join("a").join("b").join("c");
+        fs::create_dir_all(&nested).unwrap();
+
+        assert_eq!(find_git_root(&nested), Some(repo));
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn find_git_root_treats_git_file_as_repo_marker() {
+        let root = temp_root("git_root_worktree");
+        let worktree = root.join("worktree");
+        fs::create_dir_all(&worktree).unwrap();
+        fs::write(
+            worktree.join(GIT_DIR_NAME),
+            "gitdir: /elsewhere/.git/worktrees/wt\n",
+        )
+        .unwrap();
+
+        assert_eq!(find_git_root(&worktree), Some(worktree));
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn find_git_root_returns_none_when_no_git() {
+        let root = temp_root("git_root_missing");
+        let bare = root.join("bare");
+        fs::create_dir_all(&bare).unwrap();
+
+        assert_eq!(find_git_root(&bare), None);
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn bootstrap_creates_structured_layout_and_index() {
+        let root = temp_root("bootstrap_layout");
+        let repo = root.join("repo");
+        fs::create_dir_all(repo.join(GIT_DIR_NAME)).unwrap();
+
+        let mem_dir = bootstrap_workspace_memory(&repo).unwrap();
+
+        assert_eq!(mem_dir, paths::workspace_memory_dir_for(&repo));
+        assert!(mem_dir.is_dir());
+        let index = mem_dir.join(MEMORY_INDEX_FILE_NAME);
+        assert!(index.exists());
+        let body = fs::read_to_string(&index).unwrap();
+        assert!(body.starts_with("# Workspace Memory Index"));
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn bootstrap_creates_gitignore_when_absent() {
+        let root = temp_root("bootstrap_gi_new");
+        let repo = root.join("repo");
+        fs::create_dir_all(repo.join(GIT_DIR_NAME)).unwrap();
+
+        bootstrap_workspace_memory(&repo).unwrap();
+
+        let gi = repo.join(GITIGNORE_FILE_NAME);
+        assert!(gi.exists());
+        let body = fs::read_to_string(&gi).unwrap();
+        assert!(body.contains(".coyote/memory/"));
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn bootstrap_appends_to_existing_gitignore_without_trailing_newline() {
+        let root = temp_root("bootstrap_gi_append");
+        let repo = root.join("repo");
+        fs::create_dir_all(repo.join(GIT_DIR_NAME)).unwrap();
+        fs::write(repo.join(GITIGNORE_FILE_NAME), "target/").unwrap();
+
+        bootstrap_workspace_memory(&repo).unwrap();
+
+        let body = fs::read_to_string(repo.join(GITIGNORE_FILE_NAME)).unwrap();
+        assert!(body.contains("target/"));
+        assert!(body.contains(".coyote/memory/"));
+        assert!(body.ends_with('\n'));
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn bootstrap_is_idempotent_on_gitignore_entry() {
+        let root = temp_root("bootstrap_gi_idempotent");
+        let repo = root.join("repo");
+        fs::create_dir_all(repo.join(GIT_DIR_NAME)).unwrap();
+        let original = "target/\n.coyote/memory/\n";
+        fs::write(repo.join(GITIGNORE_FILE_NAME), original).unwrap();
+
+        bootstrap_workspace_memory(&repo).unwrap();
+
+        let body = fs::read_to_string(repo.join(GITIGNORE_FILE_NAME)).unwrap();
+        assert_eq!(body, original, "gitignore must be untouched");
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn bootstrap_treats_entry_without_trailing_slash_as_present() {
+        let root = temp_root("bootstrap_gi_no_slash");
+        let repo = root.join("repo");
+        fs::create_dir_all(repo.join(GIT_DIR_NAME)).unwrap();
+        let original = ".coyote/memory\n";
+        fs::write(repo.join(GITIGNORE_FILE_NAME), original).unwrap();
+
+        bootstrap_workspace_memory(&repo).unwrap();
+
+        let body = fs::read_to_string(repo.join(GITIGNORE_FILE_NAME)).unwrap();
+        assert_eq!(body, original);
+
+        let _ = fs::remove_dir_all(&root);
+    }
+
+    #[test]
+    fn bootstrap_does_not_clobber_existing_index() {
+        let root = temp_root("bootstrap_existing_index");
+        let repo = root.join("repo");
+        fs::create_dir_all(repo.join(GIT_DIR_NAME)).unwrap();
+        let mem_dir = paths::workspace_memory_dir_for(&repo);
+        fs::create_dir_all(&mem_dir).unwrap();
+        let preserved = "# Custom Index\n\n- [[foo]]: keep me\n";
+        fs::write(mem_dir.join(MEMORY_INDEX_FILE_NAME), preserved).unwrap();
+
+        bootstrap_workspace_memory(&repo).unwrap();
+
+        let body = fs::read_to_string(mem_dir.join(MEMORY_INDEX_FILE_NAME)).unwrap();
+        assert_eq!(body, preserved);
+
+        let _ = fs::remove_dir_all(&root);
+    }
+}

src/config/mod.rs

@@ -3,10 +3,12 @@ mod app_config;
 mod app_state;
 mod input;
 mod install_remote;
+pub(crate) mod instructions;
 mod macros;
 mod mcp_factory;
+pub(crate) mod memory;
 pub(crate) mod paths;
-mod prompts;
+pub(crate) mod prompts;
 mod rag_cache;
 mod request_context;
 mod role;
@@ -20,6 +22,7 @@ mod update;
 
 pub use self::agent::{
     Agent, AgentVariable, AgentVariables, complete_agent_variables, list_agents,
+    list_agents_with_descriptions,
 };
 #[allow(unused_imports)]
 pub use self::app_config::AppConfig;
@@ -28,11 +31,11 @@ pub use self::app_state::AppState;
 pub use self::input::Input;
 pub use self::install_remote::{install_remote, install_remote_from_repl_args};
 #[allow(unused_imports)]
-pub use self::request_context::{RenderMode, RequestContext};
+pub use self::request_context::{RenderMode, RequestContext, should_inject_skill_instructions};
 pub use self::role::{
     CODE_ROLE, CREATE_TITLE_ROLE, EXPLAIN_SHELL_ROLE, Role, RoleLike, SHELL_ROLE,
 };
-use self::session::Session;
+pub use self::session::Session;
 #[allow(unused_imports)]
 pub use self::skill::Skill;
 #[allow(unused_imports)]
@@ -42,7 +45,7 @@ pub use self::skill_registry::SkillRegistry;
 pub use self::update::run_self_update;
 use crate::client::{
     ClientConfig, MessageContentToolCalls, Model, ModelType, OPENAI_COMPATIBLE_PROVIDERS,
-    ProviderModels, create_client_config, list_client_types,
+    ProviderModels, create_client_config, list_client_types, oauth,
 };
 use crate::function::{FunctionDeclaration, Functions};
 use crate::rag::Rag;
@@ -61,7 +64,7 @@ use indoc::formatdoc;
 use inquire::{Confirm, Select};
 use serde::{Deserialize, Serialize};
 use serde_json::json;
-use std::collections::HashMap;
+use std::collections::{HashMap, HashSet};
 use std::sync::LazyLock;
 use std::{
     env,
@@ -134,11 +137,23 @@ const RAGS_DIR_NAME: &str = "rags";
 const FUNCTIONS_DIR_NAME: &str = "functions";
 const FUNCTIONS_BIN_DIR_NAME: &str = "bin";
 const AGENTS_DIR_NAME: &str = "agents";
+const REPL_HISTORY_DIR_NAME: &str = "repl-history";
 const GLOBAL_TOOLS_DIR_NAME: &str = "tools";
 const GLOBAL_TOOLS_UTILS_DIR_NAME: &str = "utils";
 const BASH_PROMPT_UTILS_FILE_NAME: &str = "prompt-utils.sh";
 const MCP_FILE_NAME: &str = "mcp.json";
-const DEFAULT_VISIBLE_TOOLS: [&str; 18] = [
+const HIDDEN_MCP_FILE_NAME: &str = ".mcp.json";
+const MEMORY_DIR_NAME: &str = "memory";
+const MEMORY_INDEX_FILE_NAME: &str = "MEMORY.md";
+const WORKSPACE_COYOTE_DIR_NAME: &str = ".coyote";
+const SBX_KIT_DIR_NAME: &str = "sbx-kit";
+const SBX_KIT_HASH_FILE: &str = "kit.sha256";
+const SBX_MIXIN_FILE_NAME: &str = "sbx-mixin.yaml";
+const SBX_VAULT_MIXINS_DIR_NAME: &str = "sbx-vault-mixins";
+const SBX_MIXIN_KITS_DIR_NAME: &str = "sbx-mixin-kits";
+const GIT_DIR_NAME: &str = ".git";
+const GITIGNORE_FILE_NAME: &str = ".gitignore";
+const DEFAULT_VISIBLE_TOOLS: [&str; 19] = [
     "execute_command.sh",
     "execute_py_code.py",
     "execute_sql_code.sh",
@@ -152,6 +167,7 @@ const DEFAULT_VISIBLE_TOOLS: [&str; 18] = [
     "fs_read.sh",
     "fs_rm.sh",
     "fs_write.sh",
+    "ast_grep.sh",
     "get_current_time.sh",
     "get_current_weather.sh",
     "search_wikipedia.sh",
@@ -169,7 +185,7 @@ const SUMMARIZATION_PROMPT: &str =
 const SUMMARY_CONTEXT_PROMPT: &str = "This is a summary of the chat history as a recap: ";
 
 const LEFT_PROMPT: &str = "{color.red}{model}){color.green}{?session {?agent {agent}>}{session}{?role /}}{!session {?agent {agent}>}}{role}{?rag @{rag}}{color.cyan}{?session )}{!session >}{color.reset} ";
-const RIGHT_PROMPT: &str = "{color.purple}{?session {?consume_tokens {consume_tokens}({consume_percent}%)}{!consume_tokens {consume_tokens}}}{color.reset}";
+const RIGHT_PROMPT: &str = "{color.cyan}{?reasoning_effort [{reasoning_effort}] }{color.purple}{?session {?consume_tokens {consume_tokens}({consume_percent}%)}{!consume_tokens {consume_tokens}}}{color.reset}";
 
 static EDITOR: OnceLock<Option<String>> = OnceLock::new();
 
@@ -214,6 +230,8 @@ pub struct Config {
     pub max_auto_continues: usize,
     pub inject_todo_instructions: bool,
     pub continuation_prompt: Option<String>,
+    pub inject_skill_instructions: bool,
+    pub skill_instructions: Option<String>,
 
     pub repl_prelude: Option<String>,
     pub cmd_prelude: Option<String>,
@@ -221,8 +239,17 @@ pub struct Config {
 
     pub save_session: Option<bool>,
     pub compression_threshold: usize,
+    pub compression_keep_last: usize,
     pub summarization_prompt: Option<String>,
     pub summary_context_prompt: Option<String>,
+    pub max_tool_result_chars: Option<usize>,
+
+    pub memory: Option<bool>,
+    pub memory_cap_with_tools: Option<usize>,
+    pub memory_cap_without_tools: Option<usize>,
+
+    pub workspace_instructions: Option<bool>,
+    pub workspace_instructions_files: Option<Vec<String>>,
 
     pub rag_embedding_model: Option<String>,
     pub rag_reranker_model: Option<String>,
@@ -230,11 +257,15 @@ pub struct Config {
     pub rag_chunk_size: Option<usize>,
     pub rag_chunk_overlap: Option<usize>,
     pub rag_template: Option<String>,
+    pub rag_extractor_model: Option<String>,
+    pub rag_extractor_prompt: Option<String>,
+    pub rag_graph_hops: usize,
 
     #[serde(default)]
     pub document_loaders: HashMap<String, String>,
 
     pub highlight: bool,
+    pub raw_markdown: bool,
     pub theme: Option<String>,
     pub left_prompt: Option<String>,
     pub right_prompt: Option<String>,
@@ -280,6 +311,8 @@ impl Default for Config {
             max_auto_continues: 10,
             inject_todo_instructions: true,
             continuation_prompt: None,
+            inject_skill_instructions: true,
+            skill_instructions: None,
 
             repl_prelude: None,
             cmd_prelude: None,
@@ -287,8 +320,17 @@ impl Default for Config {
 
             save_session: None,
             compression_threshold: 4000,
+            compression_keep_last: 0,
             summarization_prompt: None,
             summary_context_prompt: None,
+            max_tool_result_chars: None,
+
+            memory: None,
+            memory_cap_with_tools: None,
+            memory_cap_without_tools: None,
+
+            workspace_instructions: None,
+            workspace_instructions_files: None,
 
             rag_embedding_model: None,
             rag_reranker_model: None,
@@ -296,10 +338,14 @@ impl Default for Config {
             rag_chunk_size: None,
             rag_chunk_overlap: None,
             rag_template: None,
+            rag_extractor_model: None,
+            rag_extractor_prompt: None,
+            rag_graph_hops: 1,
 
             document_loaders: Default::default(),
 
             highlight: true,
+            raw_markdown: false,
             theme: None,
             left_prompt: None,
             right_prompt: None,
@@ -346,6 +392,12 @@ impl AssetCategory {
     }
 }
 
+#[derive(Debug, Clone, Copy, PartialEq, Eq, clap::ValueEnum)]
+pub enum MemoryScope {
+    Global,
+    Workspace,
+}
+
 #[derive(Debug, Clone, Copy, PartialEq, Eq, clap::ValueEnum)]
 pub enum InstallFilter {
     Agents,
@@ -413,9 +465,10 @@ fn confirm_asset_overwrite(category: AssetCategory, label: &str, target: &Path)
     }
     let body = match category {
         AssetCategory::McpConfig => format!(
-            "This replaces your MCP server configuration at {} with this \
+            "This merges the bundled MCP server template into your configuration \
-             build's bundled template. Your configured MCP servers (and any \
+             at {}. New servers from the bundled template will be added; any \
-             custom secret references they contain) will be lost.",
+             MCP servers you have already configured (including custom secret \
+             references) are left untouched.",
             target.display()
         ),
         _ => format!(
@@ -435,7 +488,7 @@ fn confirm_asset_overwrite(category: AssetCategory, label: &str, target: &Path)
 pub fn default_sessions_dir() -> PathBuf {
     match env::var(get_env_name("sessions_dir")) {
         Ok(value) => PathBuf::from(value),
-        Err(_) => paths::local_path(SESSIONS_DIR_NAME),
+        Err(_) => paths::local_dir(SESSIONS_DIR_NAME),
     }
 }
 
@@ -550,6 +603,18 @@ impl Config {
             })
             .with_context(|| "Failed to load config from str")?;
 
+        let mut seen = HashSet::new();
+        for cc in &config.clients {
+            let (name, _, _) = oauth::client_config_info(cc);
+            if !seen.insert(name.to_string()) {
+                bail!(
+                    "Duplicate client name '{name}' in config.yaml. \
+                     Client names must be unique across all `clients[]` entries \
+                     to avoid OAuth token collisions."
+                );
+            }
+        }
+
         Ok(config)
     }
 
@@ -642,6 +707,9 @@ bitflags::bitflags! {
         const SESSION = 1 << 2;
         const RAG = 1 << 3;
         const AGENT = 1 << 4;
+        const FUNCTION_CALLING = 1 << 5;
+        const AUTO_CONTINUE = 1 << 6;
+        const SKILLS_ENABLED = 1 << 7;
     }
 }
 

src/config/paths.rs

@@ -2,10 +2,14 @@ use super::role::Role;
 use super::{
     AGENT_GRAPH_FILE_NAME, AGENTS_DIR_NAME, BASH_PROMPT_UTILS_FILE_NAME, CONFIG_FILE_NAME,
     ENV_FILE_NAME, FUNCTIONS_BIN_DIR_NAME, FUNCTIONS_DIR_NAME, GLOBAL_TOOLS_DIR_NAME,
-    GLOBAL_TOOLS_UTILS_DIR_NAME, MACROS_DIR_NAME, MCP_FILE_NAME, ModelsOverride, RAGS_DIR_NAME,
+    GLOBAL_TOOLS_UTILS_DIR_NAME, HIDDEN_MCP_FILE_NAME, MACROS_DIR_NAME, MCP_FILE_NAME,
-    ROLES_DIR_NAME, SKILLS_DIR_NAME,
+    MEMORY_DIR_NAME, MEMORY_INDEX_FILE_NAME, ModelsOverride, RAGS_DIR_NAME, ROLES_DIR_NAME,
+    SBX_KIT_DIR_NAME, SBX_KIT_HASH_FILE, SBX_MIXIN_FILE_NAME, SBX_MIXIN_KITS_DIR_NAME,
+    SBX_VAULT_MIXINS_DIR_NAME, SKILLS_DIR_NAME, WORKSPACE_COYOTE_DIR_NAME,
 };
 use crate::client::ProviderModels;
+use crate::config::REPL_HISTORY_DIR_NAME;
+use crate::config::session::Session;
 use crate::utils::{get_env_name, list_file_names, normalize_env_name};
 
 use anyhow::{Context, Result, anyhow, bail};
@@ -13,7 +17,7 @@ use log::LevelFilter;
 use std::collections::HashSet;
 use std::env;
 use std::fs::{read_dir, read_to_string};
-use std::path::PathBuf;
+use std::path::{Path, PathBuf};
 
 pub fn config_dir() -> PathBuf {
     if let Ok(v) = env::var(get_env_name("config_dir")) {
@@ -26,38 +30,147 @@ pub fn config_dir() -> PathBuf {
     }
 }
 
-pub fn local_path(name: &str) -> PathBuf {
+pub fn local_dir(name: &str) -> PathBuf {
     config_dir().join(name)
 }
 
-pub fn cache_path() -> PathBuf {
+pub fn cache_dir() -> PathBuf {
+    if let Ok(v) = env::var(get_env_name("cache_dir")) {
+        PathBuf::from(v)
+    } else if let Ok(v) = env::var("XDG_CACHE_HOME") {
+        PathBuf::from(v).join(env!("CARGO_CRATE_NAME"))
+    } else {
         let base_dir = dirs::cache_dir().unwrap_or_else(env::temp_dir);
         base_dir.join(env!("CARGO_CRATE_NAME"))
+    }
 }
 
-pub fn oauth_tokens_path() -> PathBuf {
+pub fn sandbox_kit_override() -> Option<PathBuf> {
-    cache_path().join("oauth")
+    env::var_os(get_env_name("sandbox_kit")).map(PathBuf::from)
+}
+
+pub fn translate_sandboxed_home_dir(path: &Path) -> Option<PathBuf> {
+    env::var_os("IS_SANDBOX")?;
+
+    let s = path.to_str()?;
+
+    if let Some(translated) = translate_unix_home_style(s, "/home/") {
+        return Some(translated);
+    }
+
+    if let Some(translated) = translate_unix_home_style(s, "/Users/") {
+        return Some(translated);
+    }
+
+    translate_windows_users_dir(s)
+}
+
+fn translate_unix_home_style(s: &str, prefix: &str) -> Option<PathBuf> {
+    let rest = s.strip_prefix(prefix)?;
+    let (user, tail) = match rest.split_once('/') {
+        Some((u, t)) => (u, t),
+        None => (rest, ""),
+    };
+
+    if user.is_empty() || user == "agent" {
+        return None;
+    }
+
+    Some(if tail.is_empty() {
+        PathBuf::from("/home/agent")
+    } else {
+        PathBuf::from(format!("/home/agent/{tail}"))
+    })
+}
+
+fn translate_windows_users_dir(s: &str) -> Option<PathBuf> {
+    let bytes = s.as_bytes();
+    if bytes.len() < 4 || !bytes[0].is_ascii_alphabetic() || bytes[1] != b':' || bytes[2] != b'\\' {
+        return None;
+    }
+
+    let after_drive = &s[3..];
+    let rest = after_drive.strip_prefix("Users\\")?;
+    let (user, tail) = match rest.split_once('\\') {
+        Some((u, t)) => (u, t.replace('\\', "/")),
+        None => (rest, String::new()),
+    };
+
+    if user.is_empty() || user == "agent" {
+        return None;
+    }
+
+    Some(if tail.is_empty() {
+        PathBuf::from("/home/agent")
+    } else {
+        PathBuf::from(format!("/home/agent/{tail}"))
+    })
+}
+
+pub fn sbx_mixin_file() -> PathBuf {
+    config_dir().join(SBX_MIXIN_FILE_NAME)
+}
+
+pub fn global_tools_sbx_mixin_file() -> PathBuf {
+    functions_dir().join(SBX_MIXIN_FILE_NAME)
+}
+
+pub fn find_workspace_sbx_mixin(start: &Path) -> Option<PathBuf> {
+    for dir in start.ancestors() {
+        let candidate = dir
+            .join(WORKSPACE_COYOTE_DIR_NAME)
+            .join(SBX_MIXIN_FILE_NAME);
+        if candidate.exists() {
+            return Some(candidate);
+        }
+    }
+
+    None
+}
+
+pub fn oauth_tokens_dir() -> PathBuf {
+    cache_dir().join("oauth")
 }
 
 pub fn token_file(client_name: &str) -> PathBuf {
-    oauth_tokens_path().join(format!("{client_name}_oauth_tokens.json"))
+    oauth_tokens_dir().join(format!("{client_name}_oauth_tokens.json"))
 }
 
-pub fn log_path() -> PathBuf {
+pub fn log_file() -> PathBuf {
-    cache_path().join(format!("{}.log", env!("CARGO_CRATE_NAME")))
+    cache_dir().join(format!("{}.log", env!("CARGO_CRATE_NAME")))
+}
+
+pub fn sbx_kit_dir() -> PathBuf {
+    cache_dir().join(SBX_KIT_DIR_NAME)
+}
+
+pub fn sbx_kit_hash_file() -> PathBuf {
+    sbx_kit_dir().join(SBX_KIT_HASH_FILE)
+}
+
+pub fn sbx_vault_mixins_dir() -> PathBuf {
+    cache_dir().join(SBX_VAULT_MIXINS_DIR_NAME)
+}
+
+pub fn sbx_vault_mixins_hash_file() -> PathBuf {
+    sbx_vault_mixins_dir().join(SBX_KIT_HASH_FILE)
+}
+
+pub fn sbx_mixin_kits_dir() -> PathBuf {
+    cache_dir().join(SBX_MIXIN_KITS_DIR_NAME)
 }
 
 pub fn config_file() -> PathBuf {
     match env::var(get_env_name("config_file")) {
         Ok(value) => PathBuf::from(value),
-        Err(_) => local_path(CONFIG_FILE_NAME),
+        Err(_) => local_dir(CONFIG_FILE_NAME),
     }
 }
 
 pub fn roles_dir() -> PathBuf {
     match env::var(get_env_name("roles_dir")) {
         Ok(value) => PathBuf::from(value),
-        Err(_) => local_path(ROLES_DIR_NAME),
+        Err(_) => local_dir(ROLES_DIR_NAME),
     }
 }
 
@@ -68,7 +181,7 @@ pub fn role_file(name: &str) -> PathBuf {
 pub fn skills_dir() -> PathBuf {
     match env::var(get_env_name("skills_dir")) {
         Ok(value) => PathBuf::from(value),
-        Err(_) => local_path(SKILLS_DIR_NAME),
+        Err(_) => local_dir(SKILLS_DIR_NAME),
     }
 }
 
@@ -80,6 +193,40 @@ pub fn skill_file(name: &str) -> PathBuf {
     skill_dir(name).join("SKILL.md")
 }
 
+pub fn workspace_config_dir() -> PathBuf {
+    let workspace_dir_name = match env::var(get_env_name("workspace_config_dir")) {
+        Ok(value) => value,
+        Err(_) => WORKSPACE_COYOTE_DIR_NAME.to_string(),
+    };
+
+    env::current_dir()
+        .unwrap_or_default()
+        .join(workspace_dir_name)
+}
+
+pub fn workspace_skills_dir() -> PathBuf {
+    workspace_config_dir().join(SKILLS_DIR_NAME)
+}
+
+pub fn workspace_skill_file(name: &str) -> PathBuf {
+    workspace_skills_dir().join(name).join("SKILL.md")
+}
+
+pub fn workspace_mcp_config_file() -> Option<PathBuf> {
+    workspace_mcp_config_file_in(&env::current_dir().unwrap_or_default())
+}
+
+fn workspace_mcp_config_file_in(workspace_root: &Path) -> Option<PathBuf> {
+    let dir = workspace_config_dir();
+    [
+        dir.join(MCP_FILE_NAME),
+        dir.join(HIDDEN_MCP_FILE_NAME),
+        workspace_root.join(HIDDEN_MCP_FILE_NAME),
+    ]
+    .into_iter()
+    .find(|candidate| candidate.is_file())
+}
+
 pub fn validate_skill_name(name: &str) -> Result<()> {
     if name.is_empty() {
         bail!("Skill name cannot be empty");
@@ -96,7 +243,7 @@ pub fn validate_skill_name(name: &str) -> Result<()> {
 pub fn macros_dir() -> PathBuf {
     match env::var(get_env_name("macros_dir")) {
         Ok(value) => PathBuf::from(value),
-        Err(_) => local_path(MACROS_DIR_NAME),
+        Err(_) => local_dir(MACROS_DIR_NAME),
     }
 }
 
@@ -107,21 +254,21 @@ pub fn macro_file(name: &str) -> PathBuf {
 pub fn env_file() -> PathBuf {
     match env::var(get_env_name("env_file")) {
         Ok(value) => PathBuf::from(value),
-        Err(_) => local_path(ENV_FILE_NAME),
+        Err(_) => local_dir(ENV_FILE_NAME),
     }
 }
 
 pub fn rags_dir() -> PathBuf {
     match env::var(get_env_name("rags_dir")) {
         Ok(value) => PathBuf::from(value),
-        Err(_) => local_path(RAGS_DIR_NAME),
+        Err(_) => local_dir(RAGS_DIR_NAME),
     }
 }
 
 pub fn functions_dir() -> PathBuf {
     match env::var(get_env_name("functions_dir")) {
         Ok(value) => PathBuf::from(value),
-        Err(_) => local_path(FUNCTIONS_DIR_NAME),
+        Err(_) => local_dir(FUNCTIONS_DIR_NAME),
     }
 }
 
@@ -146,7 +293,7 @@ pub fn bash_prompt_utils_file() -> PathBuf {
 }
 
 pub fn agents_data_dir() -> PathBuf {
-    local_path(AGENTS_DIR_NAME)
+    local_dir(AGENTS_DIR_NAME)
 }
 
 pub fn agent_data_dir(name: &str) -> PathBuf {
@@ -192,7 +339,39 @@ pub fn agent_functions_file(name: &str) -> Result<PathBuf> {
 }
 
 pub fn models_override_file() -> PathBuf {
-    local_path("models-override.yaml")
+    local_dir("models-override.yaml")
+}
+
+pub fn global_memory_dir() -> PathBuf {
+    config_dir().join(MEMORY_DIR_NAME)
+}
+
+pub fn global_memory_index_file() -> PathBuf {
+    global_memory_dir().join(MEMORY_INDEX_FILE_NAME)
+}
+
+pub fn workspace_memory_dir_for(workspace_root: &Path) -> PathBuf {
+    workspace_root
+        .join(WORKSPACE_COYOTE_DIR_NAME)
+        .join(MEMORY_DIR_NAME)
+}
+
+pub fn workspace_memory_index_file_for(workspace_root: &Path) -> PathBuf {
+    workspace_memory_dir_for(workspace_root).join(MEMORY_INDEX_FILE_NAME)
+}
+
+pub fn repl_history_dir() -> PathBuf {
+    cache_dir().join(REPL_HISTORY_DIR_NAME)
+}
+
+pub fn repl_history_file(session: &Option<Session>) -> PathBuf {
+    let history_key = if let Some(session) = &session {
+        format!("session_{}", session.name().replace('/', "_"))
+    } else {
+        "default".to_string()
+    };
+
+    repl_history_dir().join(history_key)
 }
 
 pub fn log_config() -> Result<(LevelFilter, Option<PathBuf>)> {
@@ -205,7 +384,7 @@ pub fn log_config() -> Result<(LevelFilter, Option<PathBuf>)> {
         });
     let resolved_log_path = match env::var(get_env_name("log_path")) {
         Ok(v) => Some(PathBuf::from(v)),
-        Err(_) => Some(log_path()),
+        Err(_) => Some(log_file()),
     };
     Ok((log_level, resolved_log_path))
 }
@@ -264,25 +443,31 @@ pub fn has_macro(name: &str) -> bool {
 
 pub fn list_skills() -> Vec<String> {
     let mut names = Vec::new();
-    if let Ok(rd) = read_dir(skills_dir()) {
+    let mut seen = HashSet::new();
+
+    for dir in [workspace_skills_dir(), skills_dir()] {
+        if let Ok(rd) = read_dir(dir) {
             for entry in rd.flatten() {
                 if let Ok(file_type) = entry.file_type()
                     && file_type.is_dir()
                     && let Some(name) = entry.file_name().to_str()
+                    && !seen.contains(name)
                     && entry.path().join("SKILL.md").is_file()
                     && validate_skill_name(name).is_ok()
                 {
+                    seen.insert(name.to_string());
                     names.push(name.to_string());
                 }
             }
         }
+    }
 
     names.sort_unstable();
     names
 }
 
 pub fn has_skill(name: &str) -> bool {
-    skill_file(name).is_file()
+    workspace_skill_file(name).is_file() || skill_file(name).is_file()
 }
 
 pub fn local_models_override() -> Result<Vec<ProviderModels>> {
@@ -350,6 +535,295 @@ mod tests {
         }
     }
 
+    mod sandbox_home_translation {
+        use super::*;
+        use serial_test::serial;
+
+        fn with_sandbox<F: FnOnce()>(f: F) {
+            let prev = env::var_os("IS_SANDBOX");
+            unsafe {
+                env::set_var("IS_SANDBOX", "1");
+            }
+            f();
+            unsafe {
+                match prev {
+                    Some(v) => env::set_var("IS_SANDBOX", v),
+                    None => env::remove_var("IS_SANDBOX"),
+                }
+            }
+        }
+
+        fn without_sandbox<F: FnOnce()>(f: F) {
+            let prev = env::var_os("IS_SANDBOX");
+            unsafe {
+                env::remove_var("IS_SANDBOX");
+            }
+            f();
+            unsafe {
+                if let Some(v) = prev {
+                    env::set_var("IS_SANDBOX", v);
+                }
+            }
+        }
+
+        #[test]
+        #[serial]
+        fn returns_none_when_not_in_sandbox() {
+            without_sandbox(|| {
+                let p = Path::new("/home/atusa/.coyote_password");
+                assert_eq!(translate_sandboxed_home_dir(p), None);
+            });
+        }
+
+        #[test]
+        #[serial]
+        fn translates_host_home_to_agent_home() {
+            with_sandbox(|| {
+                let p = Path::new("/home/atusa/.coyote_password");
+                assert_eq!(
+                    translate_sandboxed_home_dir(p),
+                    Some(PathBuf::from("/home/agent/.coyote_password"))
+                );
+            });
+        }
+
+        #[test]
+        #[serial]
+        fn translates_nested_host_home_dir() {
+            with_sandbox(|| {
+                let p = Path::new("/home/atusa/.config/coyote/.password");
+                assert_eq!(
+                    translate_sandboxed_home_dir(p),
+                    Some(PathBuf::from("/home/agent/.config/coyote/.password"))
+                );
+            });
+        }
+
+        #[test]
+        #[serial]
+        fn returns_none_when_path_already_targets_agent_home() {
+            with_sandbox(|| {
+                let p = Path::new("/home/agent/.coyote_password");
+                assert_eq!(translate_sandboxed_home_dir(p), None);
+            });
+        }
+
+        #[test]
+        #[serial]
+        fn returns_none_when_path_is_outside_home() {
+            with_sandbox(|| {
+                let p = Path::new("/etc/coyote/.coyote_password");
+                assert_eq!(translate_sandboxed_home_dir(p), None);
+            });
+        }
+
+        #[test]
+        #[serial]
+        fn returns_none_for_relative_path() {
+            with_sandbox(|| {
+                let p = Path::new(".coyote_password");
+                assert_eq!(translate_sandboxed_home_dir(p), None);
+            });
+        }
+
+        #[test]
+        #[serial]
+        fn returns_none_for_first_segment_not_home() {
+            with_sandbox(|| {
+                let p = Path::new("/opt/atusa/.coyote_password");
+                assert_eq!(translate_sandboxed_home_dir(p), None);
+            });
+        }
+
+        #[test]
+        #[serial]
+        fn translates_macos_users_dir() {
+            with_sandbox(|| {
+                let p = Path::new("/Users/atusa/.coyote_password");
+                assert_eq!(
+                    translate_sandboxed_home_dir(p),
+                    Some(PathBuf::from("/home/agent/.coyote_password"))
+                );
+            });
+        }
+
+        #[test]
+        #[serial]
+        fn translates_macos_nested_dir() {
+            with_sandbox(|| {
+                let p = Path::new("/Users/atusa/.config/coyote/.password");
+                assert_eq!(
+                    translate_sandboxed_home_dir(p),
+                    Some(PathBuf::from("/home/agent/.config/coyote/.password"))
+                );
+            });
+        }
+
+        #[test]
+        #[serial]
+        fn returns_none_when_macos_dir_already_targets_agent() {
+            with_sandbox(|| {
+                let p = Path::new("/Users/agent/.coyote_password");
+                assert_eq!(translate_sandboxed_home_dir(p), None);
+            });
+        }
+
+        #[test]
+        #[serial]
+        fn translates_windows_drive_letter_path() {
+            with_sandbox(|| {
+                let p = Path::new("C:\\Users\\atusa\\.coyote_password");
+                assert_eq!(
+                    translate_sandboxed_home_dir(p),
+                    Some(PathBuf::from("/home/agent/.coyote_password"))
+                );
+            });
+        }
+
+        #[test]
+        #[serial]
+        fn translates_windows_nested_path() {
+            with_sandbox(|| {
+                let p = Path::new("D:\\Users\\atusa\\.config\\coyote\\.password");
+                assert_eq!(
+                    translate_sandboxed_home_dir(p),
+                    Some(PathBuf::from("/home/agent/.config/coyote/.password"))
+                );
+            });
+        }
+
+        #[test]
+        #[serial]
+        fn returns_none_when_windows_path_already_targets_agent() {
+            with_sandbox(|| {
+                let p = Path::new("C:\\Users\\agent\\.coyote_password");
+                assert_eq!(translate_sandboxed_home_dir(p), None);
+            });
+        }
+    }
+
+    mod workspace_mcp_resolution {
+        use super::*;
+        use serial_test::serial;
+
+        fn with_workspace_dir<F: FnOnce(&Path, &Path)>(f: F) {
+            let unique = time::SystemTime::now()
+                .duration_since(time::UNIX_EPOCH)
+                .unwrap()
+                .as_nanos();
+            let root = env::temp_dir().join(format!("coyote-workspace-mcp-test-{unique}"));
+            let ws_dir = root.join(WORKSPACE_COYOTE_DIR_NAME);
+            fs::create_dir_all(&ws_dir).unwrap();
+            let env_name = get_env_name("workspace_config_dir");
+            let prev = env::var_os(&env_name);
+            unsafe {
+                env::set_var(&env_name, &ws_dir);
+            }
+            f(&root, &ws_dir);
+            unsafe {
+                match prev {
+                    Some(v) => env::set_var(&env_name, v),
+                    None => env::remove_var(&env_name),
+                }
+            }
+            let _ = fs::remove_dir_all(&root);
+        }
+
+        #[test]
+        #[serial]
+        fn returns_none_when_no_config_exists() {
+            with_workspace_dir(|root, _| {
+                assert_eq!(workspace_mcp_config_file_in(root), None);
+            });
+        }
+
+        #[test]
+        #[serial]
+        fn finds_mcp_json() {
+            with_workspace_dir(|root, ws_dir| {
+                fs::write(ws_dir.join("mcp.json"), "{}").unwrap();
+                assert_eq!(
+                    workspace_mcp_config_file_in(root),
+                    Some(ws_dir.join("mcp.json"))
+                );
+            });
+        }
+
+        #[test]
+        #[serial]
+        fn falls_back_to_claude_style_hidden_mcp_json() {
+            with_workspace_dir(|root, ws_dir| {
+                fs::write(ws_dir.join(".mcp.json"), "{}").unwrap();
+                assert_eq!(
+                    workspace_mcp_config_file_in(root),
+                    Some(ws_dir.join(".mcp.json"))
+                );
+            });
+        }
+
+        #[test]
+        #[serial]
+        fn prefers_mcp_json_when_both_exist() {
+            with_workspace_dir(|root, ws_dir| {
+                fs::write(ws_dir.join("mcp.json"), "{}").unwrap();
+                fs::write(ws_dir.join(".mcp.json"), "{}").unwrap();
+                assert_eq!(
+                    workspace_mcp_config_file_in(root),
+                    Some(ws_dir.join("mcp.json"))
+                );
+            });
+        }
+
+        #[test]
+        #[serial]
+        fn falls_back_to_project_root_hidden_mcp_json() {
+            with_workspace_dir(|root, _| {
+                fs::write(root.join(".mcp.json"), "{}").unwrap();
+                assert_eq!(
+                    workspace_mcp_config_file_in(root),
+                    Some(root.join(".mcp.json"))
+                );
+            });
+        }
+
+        #[test]
+        #[serial]
+        fn prefers_workspace_dir_config_over_project_root() {
+            with_workspace_dir(|root, ws_dir| {
+                fs::write(ws_dir.join(".mcp.json"), "{}").unwrap();
+                fs::write(root.join(".mcp.json"), "{}").unwrap();
+                assert_eq!(
+                    workspace_mcp_config_file_in(root),
+                    Some(ws_dir.join(".mcp.json"))
+                );
+            });
+        }
+    }
+
+    #[test]
+    fn sandbox_kit_override_reflects_env_var_state() {
+        let env_name = get_env_name("sandbox_kit");
+        let prev = env::var_os(&env_name);
+
+        unsafe {
+            env::remove_var(&env_name);
+        }
+        assert_eq!(sandbox_kit_override(), None);
+
+        let probe = PathBuf::from("/tmp/coyote-sandbox-kit-probe");
+        unsafe {
+            env::set_var(&env_name, &probe);
+        }
+        assert_eq!(sandbox_kit_override(), Some(probe));
+
+        unsafe {
+            match prev {
+                Some(v) => env::set_var(&env_name, v),
+                None => env::remove_var(&env_name),
+            }
+        }
+    }
+
     #[test]
     fn list_skills_skips_invalid_directory_names() {
         let unique = time::SystemTime::now()

src/config/prompts.rs

@@ -1,5 +1,60 @@
 use indoc::indoc;
 
+pub(crate) const DEFAULT_SKILL_INSTRUCTIONS: &str = indoc! {"
+    ## Skills
+    Specialized skills may be available in this context. Call `skill__list` early in a task to
+    discover any that match the work, then `skill__load` the relevant ones. Their instructions and
+    granted tools will become active for subsequent turns. Call `skill__unload` when their work is
+    complete to keep the context lean."
+};
+
+pub(crate) const DEFAULT_MEMORY_INSTRUCTIONS: &str = indoc! {"
+    ## Memory
+    A persistent memory file system survives across sessions. The MEMORY.md content shown above is
+    your always-on context (universal facts, hard rules, binding feedback). Drill files hold deeper,
+    on-demand context that you fetch with `memory__read`.
+
+    Tools:
+        - `memory__read(name)`: Read a specific drill file's full content.
+        - `memory__write(name, content, scope)`: Create or replace a drill file (scope: 'global' | 'workspace').
+          The MEMORY.md index is appended automatically; do not also update the index by hand.
+          Optional `superseded_by` / `expires` (YYYY-MM-DD) mark a memory as stale for later cleanup.
+        - `memory__rename(name, new_name, scope)`: Rename a drill file. Its index entry and every
+          [[wikilink]] to it are rewritten automatically.
+        - `memory__delete(name, scope)`: Delete a drill file and its index entry. Reports any
+          [[wikilinks]] left dangling in other files.
+        - `memory__edit_index(scope, content)`: Replace the entire MEMORY.md at the given scope.
+          Use this to add always-on facts, reorganize, prune stale entries, or fix descriptions.
+        - `memory__list()`: See all known drill files and their metadata.
+        - `memory__lint()`: Health-check memory for orphans, broken links, oversized files,
+          stale (superseded/expired) files, and index descriptions that drifted from the files.
+
+    RULES:
+        - Every interaction has two outputs: your answer AND any memory updates the conversation warrants.
+          Don't let learnings evaporate into chat history.
+        - All MEMORY.md edits MUST go through `memory__edit_index`. NEVER use `fs_write`, `fs_patch`,
+          or any other generic file tool on MEMORY.md — Coyote manages its location and a stray
+          MEMORY.md outside the managed path is invisible to memory.
+        - All drill files MUST go through `memory__write`. The index updates itself. Renames and
+          deletions MUST go through `memory__rename` / `memory__delete` so links stay intact.
+        - When a fact becomes outdated, update it in place, delete it, or mark the old file with
+          `superseded_by`/`expires` so `memory__lint` flags it later. Never leave contradictory
+          memories side by side.
+        - Use [[wikilink]] notation in memory files to reference other memories by their `name:` slug.
+        - NEVER write secrets, credentials, or API keys to memory — memory is plaintext on disk.
+          Use coyote's Vault for secrets.
+        - Keep individual drill files focused (under ~2K chars). Split large topics across linked files."
+};
+
+pub(crate) const DEFAULT_MEMORY_INSTRUCTIONS_READONLY: &str = indoc! {"
+    ## Memory (read-only)
+    The memory content shown above persists across sessions. In this session it is READ-ONLY — the user
+    maintains memory files manually outside the conversation.
+
+    Reference the memory content as authoritative context about the user and their workspace.
+    Do not propose writing to memory or call any `memory__*` tools — they are unavailable."
+};
+
 pub(in crate::config) const DEFAULT_TODO_INSTRUCTIONS: &str = indoc! {"
     ## Task Tracking
     You have built-in task tracking tools. Use them to track your progress:
@@ -29,7 +84,8 @@ pub(in crate::config) const DEFAULT_SPAWN_INSTRUCTIONS: &str = indoc! {"
     | `agent__spawn` | Spawn a subagent in the background. Returns an `id` immediately. |
     | `agent__check` | Non-blocking check: is the agent done yet? Returns PENDING or result. |
     | `agent__collect` | Blocking wait: wait for an agent to finish, return its output. |
-    | `agent__list` | List all spawned agents and their status. |
+    | `agent__list_available` | List all agent types you can spawn (name + description). Use this to discover specialists before calling `agent__spawn`. |
+    | `agent__list_running` | List all subagents YOU have spawned, with their status. |
     | `agent__cancel` | Cancel a running agent by ID. |
     | `agent__task_create` | Create a task in the dependency-aware task queue. |
     | `agent__task_list` | List all tasks and their status/dependencies. |
@@ -54,6 +110,36 @@ pub(in crate::config) const DEFAULT_SPAWN_INSTRUCTIONS: &str = indoc! {"
     agent__collect --id agent_explore_e5f6g7h8
     ```
 
+    ### CRITICAL: Never end your turn with pending agents
+
+    Spawned agents do NOT report back on their own. They run in the background until you
+    actively reclaim them with `agent__collect` (to get their output) or `agent__cancel`
+    (to discard them). If you spawn agents and then emit a final message without reclaiming
+    them, the system will detect the unreclaimed agents and reject the turn-end, injecting
+    a reminder forcing you to handle them. After several such reminders, the system will
+    auto-cancel them and warn you that work was lost.
+
+    The correct flow when you have nothing else to do:
+
+    ```
+    # WRONG - do NOT do this:
+    agent__spawn --agent explore --prompt \"...\"
+    agent__spawn --agent explore --prompt \"...\"
+    # ... emit text like \"I will synthesize once they report back.\" and stop
+    # ^ The agents will be abandoned. Their output will be lost.
+
+    # RIGHT - always do this:
+    agent__spawn --agent explore --prompt \"...\"
+    agent__spawn --agent explore --prompt \"...\"
+    agent__collect --id <first_id>   # blocks until done
+    agent__collect --id <second_id>  # blocks until done
+    # ... NOW you can synthesize and end your turn
+    ```
+
+    `agent__collect` is a **blocking wait**: it pauses your execution until the agent
+    completes, then returns the output as a tool result. Use it freely — it is the
+    correct primitive for \"I'm done with my own work and just need the agents' results\".
+
     ### Parallel Spawning (DEFAULT for multi-agent work)
 
     When a task needs multiple agents, **spawn them all at once**, then collect:
@@ -121,7 +207,7 @@ pub(in crate::config) const DEFAULT_USER_INTERACTION_INSTRUCTIONS: &str = indoc!
     ## User Interaction
 
     You have built-in tools to interact with the user directly:
-        - `user__ask --question \"...\" --options [\"A\", \"B\", \"C\"]`: Present a selection prompt. Returns the chosen option.
+        - `user__select --question \"...\" --options [\"A\", \"B\", \"C\"]`: Present a single-select list of named options. Use this — not `user__confirm` — whenever there are 2+ named options. Returns the chosen option.
         - `user__confirm --question \"...\"`: Ask a yes/no question. Returns \"yes\" or \"no\".
         - `user__input --question \"...\"`: Request free-form text input from the user.
         - `user__checkbox --question \"...\" --options [\"A\", \"B\", \"C\"]`: Multi-select prompt. Returns an array of selected options.