AutoGen Security Rules for Claude Code

# AutoGen Security Rules for Claude Code

Place this in `CLAUDE.md` so Claude Code follows these conventions consistently.

## Core standards
- Keep all generated code production-ready and strongly typed where applicable.
- Prefer small focused files and functions over large mixed-responsibility modules.
- Match existing project conventions before introducing new patterns.
- Include meaningful tests for business-critical behavior.
- Never ship placeholder TODO logic in production paths.

## Security conventions
- Use explicit naming for modules, services, and handlers.
- Add boundary validation for external inputs and API payloads.
- Keep side effects isolated and observable with logs/metrics.
- Favor predictable dependency boundaries and clear ownership.
- Document trade-offs for non-obvious implementation choices.

## AutoGen guidance
- Follow canonical AutoGen project layout and idioms.
- Optimize for readability first, then measure before optimization.
- Keep configuration centralized and environment-safe.
- Ensure lint/type/test checks pass before merging.
- For python, avoid hidden magic and implicit behavior.

## Testing checklist
- Unit tests for pure logic and edge cases.
- Integration tests for external dependencies.
- Regression tests for bug fixes.
- Deterministic test data and stable assertions.

## Security & reliability checklist
- Validate and sanitize all user-controlled inputs.
- Avoid leaking secrets in logs or error responses.
- Fail safely with clear, actionable error messages.
- Add retries/timeouts only where idempotency is guaranteed.

## AI generation behavior
- Generate minimal diffs rather than full rewrites.
- Explain risky changes before applying them.