rn-code-reviewer
The reviewer agent checks implemented code for real issues with high precision. It uses confidence-based scoring to report only findings that truly matter, filtering out noise and false positives.
Overview
Section titled “Overview”| Property | Value |
|---|---|
| Model | Sonnet |
| Color | Magenta |
| Skills | rn-testing, rn-best-practices |
| Tools | Glob, Grep, LS, Read |
The reviewer is read-only. It analyzes code but does not modify it.
When triggered
Section titled “When triggered”- “review the code I just wrote for the profile edit screen”
- “check these files for any React Native specific issues before I merge”
- Phase 6 of the
rn-feature-devpipeline (3 reviewer agents check different aspects)
Confidence Scoring
Section titled “Confidence Scoring”Each potential issue is rated 0-100:
| Score | Meaning |
|---|---|
| 0 | False positive or pre-existing |
| 25 | Might be real, could be false positive |
| 50 | Real but minor or unlikely in practice |
| 75 | Verified real issue, will impact functionality |
| 100 | Confirmed definite issue, will happen frequently |
Only issues with confidence >= 80 are reported.
Review Passes
Section titled “Review Passes”Pass 1: Correctness and Bugs
Section titled “Pass 1: Correctness and Bugs”- Logic errors and undefined access paths
- Null/undefined handling in component props and state
- Race conditions in async operations (fetch + setState after unmount)
- Missing error boundaries around async data screens
- Memory leaks (uncleared intervals, uncancelled subscriptions)
Pass 2: React Native Conventions
Section titled “Pass 2: React Native Conventions”| Convention | Priority |
|---|---|
| testID coverage on all interactive elements | Critical |
__DEV__ guards on all dev-only code | Critical |
Zustand stores registered in __ZUSTAND_STORES__ | Important |
| Selector memoization (no inline .filter/.map) | Important |
| Navigation param typing | Important |
| Fast Refresh safety (no module-scope side effects) | Important |
| No bare console.log in production paths | Important |
Pass 3: Project Conventions
Section titled “Pass 3: Project Conventions”- File naming matches existing patterns
- Folder placement follows project structure
- Import style matches (relative vs alias)
- CLAUDE.md rules respected
- No duplicate code that could use existing utilities
Pass 4: Vercel RN Best Practices
Section titled “Pass 4: Vercel RN Best Practices”Applies all 46 rules from the best practices skill. Reads the full reference file for any rule category detected in the code under review.
Scanning order:
- CRITICAL (always check): falsy
&&[RN-1.1], bare strings [RN-1.2], inline components [RN-16.1] - HIGH (when keyword triggers match): list performance, animation GPU, scroll, navigation
- MEDIUM (when keyword triggers match): state, compiler, UI, design system
- LOW (only if 3+ occurrences AND confidence >= 80): monorepo, deps, JS, fonts
Output Format
Section titled “Output Format”Groups findings by severity:
Critical (confidence >= 90): clear description, file path, line number, concrete fix suggestion.
Important (confidence >= 80): same format as Critical.
If no high-confidence issues are found, confirms the code meets standards with a brief summary.