praveenmaiya/review-code

Review Code Skill

When to Use

• Before creating a PR
• User asks "review this code"
• User wants pre-merge checks

Review Checklist

Code Quality

• [ ] All functions have type hints
• [ ] Public functions have docstrings
• [ ] No hardcoded values (use configs/)
• [ ] No secrets in code
• [ ] Error handling for edge cases
• [ ] Logging instead of print statements

Tests

• [ ] Unit tests exist for new code
• [ ] Tests cover happy path and edge cases
• [ ] make test-unit passes

SQL (if applicable)

• [ ] SQL validated with make sql-validate
• [ ] Queries use parameterized inputs
• [ ] No hardcoded project/dataset names

Evaluation (if model changes)

• [ ] make eval passes
• [ ] Metrics don't regress vs baseline
• [ ] Results logged to W&B

Documentation

• [ ] agent_docs/ updated if architecture changed
• [ ] README updated if new commands/setup

Commands to Run

make lint          # Ruff + mypy
make test-unit     # Unit tests
make sql-validate  # SQL validation
make eval          # If model changes

Common Issues to Flag

1. Missing type hints - Add return types and param types
2. Hardcoded configs - Move to configs/*.yaml
3. No tests - Add tests for new functions
4. Print statements - Replace with logging
5. Magic numbers - Extract to constants or config
6. Broad exceptions - Catch specific exceptions

Generate Missing Tests

If tests are missing, suggest test cases:

def test_function_happy_path():
    """Test normal operation."""
    result = function(valid_input)
    assert result == expected

def test_function_edge_case():
    """Test edge case handling."""
    result = function(empty_input)
    assert result == default_value

def test_function_error():
    """Test error handling."""
    with pytest.raises(ValueError):
        function(invalid_input)