Code Review Expert

Perform a structured review of the current git changes with focus on SOLID principles, architecture, removal candidates, security risks, and Python-specific issues. Tailored for the Relax project (PyTorch distributed training, Ray actors, ML pipeline code).

Default to review-only output unless the user asks to implement changes.

For project structure, coupling points, and key files, see AGENTS.md.

Severity Levels

Workflow

1) Preflight Context

• Use git status -sb, git diff --stat, and git diff to scope changes.
• If needed, use rg or grep to find related modules, usages, and contracts.
• Identify entry points, ownership boundaries, and critical paths (training loop, loss computation, checkpoint saving).

Edge cases:

• No changes: Ask if they want to review staged changes or a specific commit range.
• Large diff (>500 lines): Summarize by file first, then review in batches.
• Cross-package changes: When multiple relax/ subpackages are modified, verify import compatibility first.

2) SOLID + Architecture Smells

• Load references/solid-checklist.md for specific prompts.
• When you propose a refactor, explain why it improves cohesion/coupling and outline a minimal, safe split.
• If refactor is non-trivial, propose an incremental plan instead of a large rewrite.

3) Removal Candidates + Iteration Plan

• Load references/removal-plan.md for template.
• Identify code that is unused, redundant, or feature-flagged off.
• Distinguish safe delete now vs defer with plan.

4) Security and Reliability Scan

• Load references/security-checklist.md for coverage.
• Check for: command injection, path traversal, pickle deserialization, secret leakage, race conditions, distributed race conditions.
• Call out both exploitability and impact.

5) Python-Specific Quality Scan

• Load references/code-quality-checklist.md for coverage.
• Check for: missing type hints, exception handling issues, resource management, mutable defaults, import problems.

6) ML/Training-Specific Scan

• Load references/python-ml-checklist.md for coverage.
• Check for: shape/dtype/device mismatches, gradient issues, memory leaks, distributed training bugs, numerical stability.

7) Output Format

## Code Review Summary

**Files reviewed**: X files, Y lines changed
**Overall assessment**: [APPROVE / REQUEST_CHANGES / COMMENT]

---

## Findings

### P0 - Critical
(none or list)

### P1 - High
1. **[file:line]** Brief title
   - Description of issue
   - Suggested fix

### P2 - Medium
...

### P3 - Low
...

---

## Removal/Iteration Plan
(if applicable)

Clean review: If no issues found, state what was checked and any residual risks.

8) Next Steps Confirmation

After presenting findings, ask user how to proceed:

1. Fix all - Implement all suggested fixes
2. Fix P0/P1 only - Address critical and high priority issues
3. Fix specific items - User specifies which issues to fix
4. No changes - Review complete

Important: Do NOT implement any changes until user explicitly confirms.

Resources