This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
Claude Code AutoPilot is a sophisticated plugin that enables 24-hour fully autonomous project execution through pure slash command implementation. It's a zero-dependency, lightweight system focused on requirement alignment and autonomous execution.
- Zero External Dependencies: Only relies on Claude Code and basic shell commands
- Global Marketplace Installation: Installation through local marketplace following official standards
- Slash Command Driven: All functionality through standard slash commands
- JSON State Management: Five core state files track the entire project lifecycle
- Deep Requirement Discussion: Structured analysis of user needs and decision points
- Execution Plan Generation: Dynamic task decomposition and state file creation
- Autonomous Execution: Continuous task execution with real-time status updates
/autopilot-start- Initiates the complete workflow from requirements to autonomous execution/autopilot-status- Shows current execution progress, alignment, and system health/autopilot-continue- Intelligent task execution with automatic recovery and context management/autopilot-help- Displays usage guidance and troubleshooting information
- Simplified Design: Only 4 commands to remember, reducing cognitive load
- Intelligent Automation: Automatic state detection, recovery, and context management
- Unified Functionality: Each command combines multiple related capabilities
- Professional Agents: Automatically activates appropriate expert agents based on task needs
# Clone the repository (one time)
git clone https://github.com/x-rush/claude-code-autopilot.git
# Start Claude Code
claude --dangerously-skip-permissions
# Add the plugin directory as a marketplace
/plugin marketplace add /path/to/claude-code-autopilot
# Install the plugin
/plugin install claude-code-autopilot@claude-code-autopilot# Example with absolute path
/plugin marketplace add ~/dev/claude-code-autopilot
/plugin install claude-code-autopilot@claude-code-autopilot# Check plugin installation
/help | grep autopilot
# Start the workflow
/autopilot-startThe plugin uses five state files that are generated dynamically based on templates:
- REQUIREMENT_ALIGNMENT.json - Requirements validation and alignment
- EXECUTION_PLAN.json - Task breakdown and execution strategy
- TODO_TRACKER.json - Task progress tracking and quality metrics
- DECISION_LOG.json - Decision-making process documentation
- EXECUTION_STATE.json - Current execution status and session information
skills/*/templates/*.json- Structure definitions accessed via @-file referencesskills/*/examples/*.json- Usage examples accessed via @-file references- Project root
*.json- Runtime-generated project-specific state files (git-ignored)
claude-code-autopilot/
├── .claude-plugin/plugin.json # Plugin metadata
├── commands/ # 4 simplified slash commands
│ ├── autopilot-start.md # Unified entry point
│ ├── autopilot-status.md # Multi-functional status
│ ├── autopilot-continue.md # Intelligent execution
│ └── autopilot-help.md # Help and guidance
├── agents/ # Professional expert agents
│ ├── requirement-analyst.md # Requirements analysis
│ ├── execution-planner.md # Task planning
│ └── quality-assurance.md # Quality control
├── skills/ # Modular skill systems
│ ├── requirement-alignment/ # Requirements skill
│ ├── execution-planning/ # Planning skill
│ └── state-management/ # State management skill
└── docs/ # Documentation
When modifying or debugging the plugin:
- Template Files: These define the structure that Claude Code uses to generate runtime files
- Runtime Files: Generated dynamically based on user requirements discussions
- State Consistency: All five state files must maintain data consistency
- Tasks are scored 0-10 for quality tracking
- Multi-stage validation: Requirements → Planning → Execution → Quality
- Continuous alignment checking with original requirements
- Follow the zero-dependency philosophy strictly
- Maintain the three-phase workflow constraint
- Ensure template files are well-documented with clear field descriptions
- Test command implementations thoroughly
- Always start with
/autopilot-startfor new projects - Monitor progress with
/autopilot-statusregularly - Use
/autopilot-status --alignmentto verify requirement alignment - Use
/autopilot-continue --recoverfor error recovery
- Operations are restricted to the current project directory
- No system-level commands or privileged operations
- All state management through safe JSON file operations
- Complete audit trail of all decisions and actions
- Plugin Loading: Verify installation with
/help | grep autopilot - State File Issues: Use
/autopilot-continue --recoverfor automatic repairs - Context Limits: Use
/autopilot-continue --refreshfor automatic context management - Execution Interruption: Check status with
/autopilot-statusand recover as needed
# Quick status check
/autopilot-status --quick
# Check specific aspects
/autopilot-status --aspect=requirements
/autopilot-status --aspect=execution
# Recovery modes
/autopilot-recovery --check
/autopilot-recovery --auto-fix
/autopilot-recovery --interactive- Template files in
templates/are versioned - Runtime state files in project root are git-ignored
- Local marketplace files in
local-marketplace/can be versioned with project
- All file operations are limited to the project directory
- State files are automatically managed by Claude Code
- Manual editing of state files should be done carefully
When extending the plugin:
- Follow existing command patterns in
commands/ - Update relevant JSON templates if adding new state tracking
- Maintain compatibility with the three-phase workflow
- Ensure proper error handling and recovery mechanisms
- The plugin is designed to work standalone
- External integrations should be handled through the decision framework
- All external tool usage must be explicitly planned and tracked
- Current Version: v1.0.0 (released 2025-10-17)
- Claude Code Requirement: >=1.0.0
- License: MIT
- Zero Dependency: Pure Claude Code capability utilization
- Official Standards Compliance: Installation through local marketplace mechanism
- Project Autonomy: Each project manages its own instance and marketplace
- Workflow Constraint: Strict three-phase execution model
- State Persistence: Complete execution tracking through JSON files
- Continuous Alignment: Ongoing verification with original requirements
- Autonomous Recovery: Intelligent error detection and recovery
This plugin represents a sophisticated approach to autonomous project execution that maintains strict alignment with user requirements while providing complete transparency and control through its comprehensive state management system.