This file defines what Claude Code is allowed to modify when performing automated improvements.
Claude Code MAY modify:
- Documentation files (README.md, docs/*, *.md)
- Source code for:
- Bug fixes and error handling
- Code quality improvements (remove duplication, add type hints)
- Performance optimizations
- Refactoring for clarity/maintainability
- Security fixes
- Test files (add new tests, fix broken tests, improve coverage)
- CI/CD configurations (.github/workflows/*, .gitlab-ci.yml, etc.)
- Configuration files (build configs, linting rules, etc.)
- Metadata files (package.json, pyproject.toml, setup.py)
- GitHub-specific files (.gitignore, CODEOWNERS, etc.)
- Dependencies (with clear justification for updates/additions)
Claude Code MUST NOT:
- Break public APIs or change function signatures without discussion
- Make breaking changes to existing behavior
- Change core algorithms without explaining rationale
- Remove features without justification
- Add heavy dependencies without clear need
- Modify production configuration (database URLs, API keys, etc.)
When an issue is labeled ai-implement, Claude Code may modify source code to implement the requested feature or fix, but must:
- Reference the issue number in the PR
- Follow existing code style and patterns
- Include tests if the project has a test suite
- Not change public APIs without discussion in the issue
All PRs created by Claude Code MUST include:
- A clear description of what was changed and why
- A receipt line in the format:
Receipt: <hash> - Reference to the workflow run that created it
- Minimal, focused changes (prefer small PRs over large ones)
- Minimal Deltas: Make the smallest change that solves the problem
- No Breaking Changes: Never introduce breaking changes
- Follow Existing Patterns: Match the style and structure of existing code
- Documentation First: When in doubt, improve documentation rather than code
- Safety First: If uncertain, create an issue for human review instead of making changes
This file was created to ensure safe and predictable AI-assisted repository maintenance.