review-changes

name	review-changes
description	Dispatch 5 parallel review agents to analyze pending changes
context	fork
user-invocable	true

Review Changes

Dispatch five specialized review agents in parallel to analyze uncommitted or staged changes.

When to Use

Run before committing significant changes, after completing a feature, or when requested. Particularly valuable for changes touching FFI boundaries or multiple modules.

Workflow

1. Gather Context

git diff --stat                    # What files changed
git diff                           # Full diff
git diff --cached                  # Staged changes

2. Dispatch Five Review Agents (Parallel)

All five agents run simultaneously on the diff output:

Security Reviewer

• Scans for unsafe blocks without // SAFETY: comments
• Checks FFI functions for null pointer validation
• Looks for hardcoded secrets, API keys, credentials
• Validates memory ownership documentation on pointer parameters
• Flags new external crate additions for review

Performance Reviewer

• Identifies unnecessary allocations (.clone(), String where &str suffices)
• Checks for missing #[inline] on hot-path functions
• Looks for O(n²) patterns in loops
• Flags large stack allocations in FFI boundary code
• Reviews SpriteBatch usage for unnecessary draw calls

Architecture Reviewer

• Validates dependency flow (layers flow DOWN only)
• Checks for cross-module boundary violations
• Verifies new files are in correct module directories
• Ensures no upward imports in the 5-layer hierarchy:
  • Layer 1 (Core): libs/
  • Layer 2 (Engine): goud_engine/src/
  • Layer 3 (FFI): goud_engine/src/ffi/
  • Layer 4 (SDKs): sdks/
  • Layer 5 (Apps): examples/

Pattern Reviewer

• Checks against the 16 anti-patterns from AGENTS.md
• Validates FFI patterns (#[no_mangle], #[repr(C)])
• Ensures SDK wrappers remain thin (no logic)
• Verifies error handling uses Result, not unwrap/expect in library code
• Checks component patterns (derive Debug + Clone)

Simplicity Reviewer

• Flags files exceeding 500 lines
• Identifies over-engineered abstractions
• Checks for dead code or unused imports
• Looks for duplicated logic between Rust and SDK code
• Suggests simplifications

3. Aggregate Results

Collect findings from all five agents and categorize:

Priority	Meaning	Action
P1 - BLOCKER	Must fix before commit	Blocks merge
P2 - WARNING	Should fix, strong recommendation	Review before merge
P3 - NOTE	Minor improvement suggestion	Optional

4. Output Format

## Review Summary

**Verdict**: APPROVED | CHANGES REQUESTED | BLOCKED

### BLOCKERS (P1)
| # | Reviewer | File | Finding |
|---|----------|------|---------|

### WARNINGS (P2)
| # | Reviewer | File | Finding |
|---|----------|------|---------|

### NOTES (P3)
| # | Reviewer | File | Finding |
|---|----------|------|---------|

When to Block

A review returns BLOCKED if any P1 findings exist:

• unsafe without SAFETY comment
• Secrets in code
• Upward dependency violation
• SDK logic that should be in Rust
• Missing FFI null checks on pointer parameters