Part II—Comprehensive Checklist
The core of the AI reviewer is a reproducible multi-dimensional checklist.
Functional Correctness & Logic
-
Logic contradictions, boundaries/exceptions, error handling, race conditions
-
Alignment with Intent (Consistency with PR description/ticket)
-
Not passive but Adversarial Test Generation (Mutation / Symbolic execution)
Maintainability, Readability, Complexity
-
Cyclomatic/Cognitive complexity, Code smells, Naming, Comment quality
-
Handle style guides by machine to avoid subjective debates
Architecture Fit
- Principles like SOLID, Layering, Dependency direction, Separation of concerns
Implementation Note: Connect with the Add a new skill guide in this repository, and declare check items in machine-readable (e.g., YAML) form -> Evaluate/Aggregate in CI.