Autonomous Setup

# Autonomous Setup

Configure autonomous development mode for your project.

## Context
$ARGUMENTS

## Execute

### Step 1: Check Prerequisites

```bash
# Check GitHub CLI
gh --version

# Check Node.js (for headless mode)
node --version

# Check if in a git repo
git status

Required:

• GitHub CLI authenticated (gh auth login)
• Git repository initialized

For headless mode (Ralph):

• Node.js 18+
• ANTHROPIC_API_KEY environment variable

Step 2: Create Configuration

If .claude/autonomous.yml doesn't exist, create it from template:

# Check for existing config
ls .claude/autonomous.yml 2>/dev/null || echo "No config found"

Default configuration (copy to .claude/autonomous.yml):

# Autonomous Development Configuration
target_score: 95
max_iterations: 15

gates:
  test:     { weight: 25, required: true,  auto_fix: true  }
  security: { weight: 25, required: true,  auto_fix: true  }
  build:    { weight: 15, required: true,  auto_fix: true  }
  review:   { weight: 20, required: false, auto_fix: true  }
  mentor:   { weight: 10, required: false, auto_fix: false }
  ux:       { weight: 5,  required: false, auto_fix: false }

presets:
  prototype:
    target_score: 80
  production:
    target_score: 98

Step 3: MCP Configuration

Check for GitHub MCP (required for issue tracking):

# Check MCP config
cat .mcp.json 2>/dev/null | grep -i github || echo "GitHub MCP not configured"

If not configured, recommend running /mcp-setup first.

Step 4: Explain the Two Modes

Interactive Mode (/autonomous):

• Runs inside your Claude Code session
• You can watch progress in real-time
• Intervene with @claude comments
• Best for single issues while working

# Example
/autonomous --issue 123
/autonomous --issue 123 --preset production

Headless Mode (Ralph):

• Runs outside Claude, uses Agent SDK
• Unattended overnight/CI runs
• Docker support for containers
• Best for batch processing

# Install dependencies
cd scripts && npm install

# Run headless
npx ts-node ralph-runner.ts --issue=123
npx ts-node ralph-runner.ts --issue=123 --preset=production

# Docker
docker build -t ralph-wiggum scripts/
docker run -e ANTHROPIC_API_KEY=$ANTHROPIC_API_KEY ralph-wiggum --issue=123

Step 5: Create Your First Issue

For autonomous development to work, you need properly structured issues:

## Acceptance Criteria
- [ ] User can login with email/password
- [ ] Failed logins are rate limited
- [ ] Session expires after 24 hours

## Technical Notes
- Use bcrypt for password hashing
- Store sessions in Redis

Create via /plan:

/plan "User authentication" --issues

Step 6: Verify Setup

Run a dry-run to verify everything works:

# Interactive check
/autonomous --issue 123 --dry-run

# Headless check
cd scripts && npx ts-node ralph-runner.ts --issue=123 --dry-run

Output Format

## Autonomous Setup Complete

### Prerequisites
| Requirement | Status |
|-------------|--------|
| GitHub CLI | [✅/❌] |
| Git repo | [✅/❌] |
| Node.js 18+ | [✅/❌] (headless only) |
| ANTHROPIC_API_KEY | [✅/❌] (headless only) |

### Configuration
- Config file: [created/exists/missing]
- Target score: [95]
- Presets available: [default, prototype, production]

### MCP Status
| MCP | Status |
|-----|--------|
| GitHub | [✅/❌] Required for issue tracking |

### Quick Start

**Interactive (in Claude Code):**
```bash
/autonomous --issue YOUR_ISSUE_NUMBER

Headless (background):

cd scripts && npx ts-node ralph-runner.ts --issue=YOUR_ISSUE_NUMBER

Next Steps

1. Create an issue with acceptance criteria
2. Run /autonomous --issue NUMBER --dry-run to test
3. Run /autonomous --issue NUMBER to start

Documentation

• Skill: skills/workflow/autonomous-development/SKILL.md
• Config template: templates/autonomous.yml
• Headless runner: scripts/README.md

## Troubleshooting

| Issue | Solution |
|-------|----------|
| "GitHub MCP not found" | Run `/mcp-setup` and enable GitHub |
| "gh: command not found" | Install GitHub CLI: `brew install gh` |
| "ANTHROPIC_API_KEY not set" | `export ANTHROPIC_API_KEY=your-key` |
| "No issues found" | Create issues with `/plan --issues` |
| "Score stuck" | Check for flaky tests, review gate output |

Start setup now.