version: "1.0.1" name: auto-claude description: Autonomous multi-agent coding with git worktree isolation, QA validation, and memory. Use for complex features requiring autonomous implementation. disable-model-invocation: true

Auto-Claude: Autonomous Coding Framework

Autonomous multi-agent system for complex feature implementation with QA validation.

When to Use

✅ Use Auto-Claude for:

• Complexity 3-4 - Well-defined features, multiple files
• Established codebases with clear patterns
• Repetitive tasks (CRUD, forms, API endpoints, auth flows)
• When you prefer review-at-end vs step-by-step
• Features requiring iterative refinement with QA validation

❌ Don't use Auto-Claude for:

• Complexity 1-2 - Use manual /start-task instead
• Complexity 5+ - Use /gsd:new-project instead
• Greenfield projects (no existing codebase patterns)
• Exploratory/research tasks (unclear requirements)
• Simple typo fixes or config changes
• Tasks requiring human decision-making during implementation

Quick Guide:

• Complexity 1-2 → Manual implementation
• Complexity 3-4 → Auto-Claude (autonomous)
• Complexity 5+ → GSD (multi-phase)

Core Capabilities

1. Codebase Analysis - Generates semantic baseline
2. Spec Generation - Creates detailed implementation plan
3. Autonomous Implementation - Multi-agent coordination
4. Git Worktree Isolation - Safe development in separate worktree
5. QA Validation - Iterative quality checks
6. Memory System - Graphiti-powered context retention

Workflow

# From project directory:
/auto-claude Add user authentication with JWT tokens

What Happens:

1. Analyzes project structure → generates baseline
2. Creates implementation spec → stores in ~/.auto-claude/specs/
3. Creates git worktree → isolated development environment
4. Autonomous agents implement → with QA loops
5. Returns completed code → ready for review

Output Location: ~/.auto-claude/ (gitignored)

Integration with Existing Workflows

Manual Implementation:

/start-task [description]  → Claude implements step-by-step

Autonomous Implementation:

/auto-claude [description] → Auto-Claude implements autonomously

Multi-Phase Projects:

/gsd:new-project          → GSD for project management

Configuration

Location: ~/.auto-claude/.env

Required:

• CLAUDE_CODE_OAUTH_TOKEN - Get from /settings in Claude Code

Optional (Recommended):

• GRAPHITI_ENABLED=true - Semantic memory
• GOOGLE_API_KEY - For Gemini embeddings (memory feature)
• Alternative: Ollama (local) or disable memory

Trade-offs

Strengths:

• Autonomous with minimal oversight
• QA validation ensures quality
• Git worktree prevents main branch disruption
• Memory learns from previous sessions

Limitations:

• Best for established codebases
• Requires clear specifications
• Uses tokens for multi-agent iterations
• Manual merge required after completion

Examples

Good use cases:

• "Add user authentication with JWT tokens and session management"
• "Implement dark mode toggle with localStorage persistence"
• "Refactor API layer to use async/await instead of callbacks"
• "Add pagination to user list with 20 items per page"
• "Create export-to-CSV functionality for reports"

Poor use cases:

• "Fix typo in README" (too simple)
• "Explore best approach for state management" (exploratory)
• "Build entire e-commerce platform" (too large, use GSD)
• "Make the app better" (too vague)

Typical Timeline

• Analysis: 30 seconds
• Spec Generation: 1-2 minutes
• Implementation: 5-15 minutes (depends on complexity)
• QA Validation: 2-5 minutes
• Your Review: 5-10 minutes

Total: 15-30 minutes for most features

Related

• generic-feature-developer - Manual implementation patterns
• debug-systematic - Systematic debugging
• test-specialist - Testing strategies

Documentation

See ~/.claude/docs/AUTO-CLAUDE-GUIDE.md for:

• Complete setup instructions
• Real-world examples with detailed walkthroughs
• Troubleshooting guide
• Performance tips
• Comparison with other workflows