Configuration

Customize the AI Agent Starter for your project.

CLAUDE.md

Main configuration file. Controls:

• Available commands
• Active skills
• Workflow preferences
• Output style

# CLAUDE.md

## Slash Commands
- `/plan` - Create implementation plans
- `/code` - Execute plans
...

Output Styles

Control response verbosity in .claude/output-styles/:

Level	File	Audience
0	coding-level-0-eli5.md	Non-technical
1	coding-level-1-junior.md	Junior devs
2	coding-level-2-mid.md	Mid-level
3	coding-level-3-senior.md	Senior (default)
4	coding-level-4-lead.md	Tech leads
5	coding-level-5-god.md	Experts

To change default, reference the style in your prompt or CLAUDE.md.

Hooks

Configure hooks in .claude/hooks/:

// .claude/hooks/pre-commit.js
module.exports = async (context) => {
  // Run before commits
}

See Hooks Reference for details.

Skills Activation

Skills auto-activate based on context. To force activation:

Use the debugging skill to analyze this error

Available skills:

• planning - Plan creation and validation
• code-review - Review protocols
• debugging - Root cause analysis
• pair-programming - Collaborative modes
• ai-multimodal - Image/video processing
• media-processing - FFmpeg/ImageMagick

Plans Directory

Plans stored in ./plans/:

plans/
├── {YYMMDD-HHMM}-{slug}/
│   ├── plan.md
│   ├── phase-01-*.md
│   └── research/
└── reports/

Naming convention: Date-time prefix + descriptive slug.

Environment Variables

Variable	Purpose
GEMINI_API_KEY	AI Multimodal features
ANTHROPIC_API_KEY	Claude API access
GITHUB_TOKEN	GitHub CLI operations

Security

Store secrets in .env and verify it's in .gitignore:

# Check .gitignore includes .env
grep -q "^\.env$" .gitignore || echo ".env" >> .gitignore