code-reviewer

name	code-reviewer
description	Code review and refactoring guidelines

Code Reviewer Skill

Before Making Changes

1. Read the code first - Never change code you haven't read
2. Run tests before changes - Establish baseline
3. Understand the context - Why does this code exist?

Review Checklist

Security

• No patient data logged to console/files
• No hardcoded secrets or API keys
• Environment variables for sensitive config
• Input validation on API boundaries

Async/Await

• All I/O operations use async/await
• No blocking calls in async functions
• Proper exception handling in async code

Type Safety

• All function signatures have type hints
• Pydantic models for data validation
• No Any types without justification

Testing

• New code has corresponding tests
• Tests mock external services (LLM, DB)
• Tests run fast (< 5 seconds for unit tests)

API Changes

• Check for breaking changes
• Version bump if breaking
• Update API documentation

Healthcare Specifics

• Flags have source: "rule" | "ai" tag
• LLM output validated against Pydantic models
• Audit logging for AI decisions

Red Flags to Watch For

# Bad: Hardcoded secret
api_key = "sk-xxx..."

# Bad: Patient data in logs
print(f"Processing patient: {patient.name}")

# Bad: Missing await
def get_patient(id):  # Should be async
    return db.query(...)

# Bad: Pydantic v1 pattern
obj = Model.parse_obj(data)  # Use model_validate()

# Bad: Direct anthropic import
from anthropic import Anthropic  # Use claude_agent_sdk

After Making Changes

1. Run tests after changes - Compare with baseline
2. Run linter - uv run ruff check .
3. Check type errors - If using mypy/pyright
4. Manual smoke test - Does the app still work?

Refactoring Safety

• Make small, incremental changes
• Run tests after each change
• Keep commits atomic (one logical change per commit)
• If unsure, ask before refactoring