gemini-cli-extensions · bardusco · Dec 23, 2025 · Dec 23, 2025 · Dec 23, 2025 · Dec 23, 2025
diff --git a/README.md b/README.md
@@ -19,6 +19,8 @@ The philosophy behind Conductor is simple: control your code. By treating contex
 
 ## Installation
 
+### Gemini CLI Extension
+
 Install the Conductor extension by running the following command from your terminal:
 
 ```bash
@@ -27,6 +29,28 @@ gemini extensions install https://github.com/gemini-cli-extensions/conductor --a
 
 The `--auto-update` is optional: if specified, it will update to new versions as they are released.
 
+### AI Agent Skill (Claude CLI / OpenCode / Codex)
+
+Conductor is also available as a portable **skill** that works with Claude CLI, OpenCode, and other AI coding agents that support the skills standard.
+
+```bash
+# Clone the repository
+git clone https://github.com/gemini-cli-extensions/conductor.git
+cd conductor
+
+# Run the install script
+./skill/scripts/install.sh
+```
+
+The installer will ask where to install:
+1. **OpenCode global** (`~/.opencode/skill/conductor/`)
+2. **Claude CLI global** (`~/.claude/skills/conductor/`)
+3. **Both**
+
+The skill is installed with symlinks to this repository, so running `git pull` will automatically update the skill.
+
+After installation, restart your AI CLI. The agent will automatically detect and use Conductor when you ask to create a new feature, write a spec, plan a feature, or set up a project.
+
 ## Usage
 
 Conductor is designed to manage the entire lifecycle of your development tasks.

diff --git a/skill/SKILL.md b/skill/SKILL.md
@@ -0,0 +1,94 @@
+---
+name: conductor
+description: Use when the user wants to setup a new project, create a new feature, write a spec, plan a feature, fix a bug with a plan, start a new track, check project status, implement next task, or revert changes. Also use when user mentions "conductor", "track", or "spec-driven development". If conductor is not yet configured in the project, start with setup.
+---
+
+# Conductor
+
+Conductor is a Context-Driven Development (CDD) framework that transforms AI agents into proactive project managers. The philosophy is "Measure twice, code once" - every feature follows a strict protocol: **Context -> Spec & Plan -> Implement**.
+
+## Core Concepts
+
+- **Track**: A unit of work (feature or bug fix) with its own spec and plan
+- **Spec**: Detailed requirements document (`spec.md`)
+- **Plan**: Phased task list with checkboxes (`plan.md`)
+- **Workflow**: Rules for task lifecycle, TDD, commits, and quality gates
+
+## Directory Structure
+
+When initialized, Conductor creates this structure in the project:
+
+```
+conductor/
+├── product.md              # Product vision and goals
+├── product-guidelines.md   # UX/brand guidelines
+├── tech-stack.md           # Technology choices
+├── workflow.md             # Development workflow rules
+├── tracks.md               # Master list of all tracks
+├── code_styleguides/       # Language-specific style guides
+├── tracks/                 # Active tracks
+│   └── <track_id>/
+│       ├── metadata.json
+│       ├── spec.md
+│       └── plan.md
+└── archive/                # Completed tracks
+```
+
+## Available Commands
+
+| Command | Purpose |
+|---------|---------|
+| **Setup** | Initialize Conductor in a project (new or existing) |
+| **New Track** | Create a new feature/bug track with spec and plan |
+| **Implement** | Execute tasks from a track's plan following TDD workflow |
+| **Status** | Show progress overview of all tracks |
+| **Revert** | Git-aware rollback of tracks, phases, or tasks |
+
+## Protocol References
+
+The detailed protocols are in TOML format. Read the `prompt` field from each file:
+
+| Action | Protocol File |
+|--------|---------------|
+| Setup project | `commands/conductor/setup.toml` |
+| Create new track | `commands/conductor/newTrack.toml` |
+| Implement tasks | `commands/conductor/implement.toml` |
+| Check status | `commands/conductor/status.toml` |
+| Revert changes | `commands/conductor/revert.toml` |
+
+**How to read**: Each `.toml` file has a `prompt` field containing the full protocol instructions.
+
+## Task Status Markers
+
+- `[ ]` - Pending
+- `[~]` - In Progress
+- `[x]` - Completed
+
+## Key Workflow Principles
+
+1. **The Plan is Source of Truth**: All work tracked in `plan.md`
+2. **Test-Driven Development**: Write tests before implementing
+3. **High Code Coverage**: Target >80% coverage
+4. **Commit After Each Task**: With git notes for traceability
+5. **Phase Checkpoints**: Manual verification at phase completion
+
+## When to Use Each Protocol
+
+- **"set up conductor" or "initialize project"** -> Read `commands/conductor/setup.toml`
+- **"new feature", "new track", "plan a feature"** -> Read `commands/conductor/newTrack.toml`
+- **"implement", "start working", "next task"** -> Read `commands/conductor/implement.toml`
+- **"status", "progress", "where are we"** -> Read `commands/conductor/status.toml`
+- **"revert", "undo", "rollback"** -> Read `commands/conductor/revert.toml`
+
+## Assets
+
+- **Code Styleguides**: `templates/code_styleguides/` (general, go, python, javascript, typescript, html-css)
+- **Workflow Template**: `templates/workflow.md`
+
+## Critical Rules
+
+1. **Validate every tool call** - If any fails, halt and report to user
+2. **Sequential questions** - Ask one question at a time, wait for response
+3. **User confirmation required** - Before writing files or making changes
+4. **Check setup first** - Verify `conductor/` exists before any operation
+5. **Agnostic language** - Do not suggest slash commands like `/conductor:xxx`. Instead, tell the user to ask you directly (e.g., "to start implementing, just ask me" instead of "run /conductor:implement")
diff --git a/skill/scripts/install.sh b/skill/scripts/install.sh
@@ -0,0 +1,89 @@
+#!/bin/bash
+# Install Conductor skill for Claude CLI / OpenCode
+# Usage: ./install.sh
+#
+# This script creates a skill directory with symlinks to the Conductor repository,
+# so updates to the repo are automatically reflected in the skill.
+
+set -e
+
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+SKILL_DIR="$(dirname "$SCRIPT_DIR")"
+CONDUCTOR_ROOT="$(dirname "$SKILL_DIR")"
+
+echo "Conductor Skill Installer"
+echo "========================="
+echo ""
+
+# Check if we're running from within a conductor repo
+if [ ! -f "$CONDUCTOR_ROOT/commands/conductor/setup.toml" ]; then
+    echo "Error: This script must be run from within the Conductor repository."
+    echo "Expected to find: $CONDUCTOR_ROOT/commands/conductor/setup.toml"
+    echo ""
+    echo "Please clone the repository first:"
+    echo "  git clone https://github.com/gemini-cli-extensions/conductor.git"
+    echo "  cd conductor"
+    echo "  ./skill/scripts/install.sh"
+    exit 1
+fi
+
+echo "Conductor repository found at: $CONDUCTOR_ROOT"
+echo ""
+echo "Where do you want to install the skill?"
+echo ""
+echo "  1) OpenCode global    (~/.opencode/skill/conductor/)"
+echo "  2) Claude CLI global  (~/.claude/skills/conductor/)"
+echo "  3) Both"
+echo ""
+read -p "Choose [1/2/3]: " choice
+
+case "$choice" in
+    1)
+        TARGETS=("$HOME/.opencode/skill/conductor")
+        ;;
+    2)
+        TARGETS=("$HOME/.claude/skills/conductor")
+        ;;
+    3)
+        TARGETS=("$HOME/.opencode/skill/conductor" "$HOME/.claude/skills/conductor")
+        ;;
+    *)
+        echo "Invalid choice. Exiting."
+        exit 1
+        ;;
+esac
+
+for TARGET_DIR in "${TARGETS[@]}"; do
+    echo ""
+    echo "Installing to: $TARGET_DIR"
+
+    # Remove existing installation
+    rm -rf "$TARGET_DIR"
+
+    # Create skill directory
+    mkdir -p "$TARGET_DIR"
+
+    # Copy SKILL.md (the only actual file)
+    cp "$SKILL_DIR/SKILL.md" "$TARGET_DIR/"
+
+    # Create symlinks to conductor repo directories
+    ln -s "$CONDUCTOR_ROOT/commands" "$TARGET_DIR/commands"
+    ln -s "$CONDUCTOR_ROOT/templates" "$TARGET_DIR/templates"
+
+    echo "  Created: $TARGET_DIR/SKILL.md"
+    echo "  Symlink: $TARGET_DIR/commands -> $CONDUCTOR_ROOT/commands"
+    echo "  Symlink: $TARGET_DIR/templates -> $CONDUCTOR_ROOT/templates"
+done
+
+echo ""
+echo "Conductor skill installed successfully!"
+echo ""
+echo "Structure:"
+for TARGET_DIR in "${TARGETS[@]}"; do
+    ls -la "$TARGET_DIR" 2>/dev/null || true
+done
+echo ""
+echo "The skill references the Conductor repo at: $CONDUCTOR_ROOT"
+echo "Updates to the repo (git pull) will be reflected automatically."
+echo ""
+echo "Restart your AI CLI to load the skill."