cr-react

name	cr-react
description	Run a 7-pass React/Next.js code review scoped to the current PR diff. Use when reviewing pull requests, checking code before merge, or when the user asks for a React code review.
disable-model-invocation	true

CR-React — Diff-Scoped React Review

Review scope: PR diff + directly affected files only.

Follow the review criteria defined in rules/react.md.

Step 1 — Detect project context

Read package.json and check for:

• next dependency → enable SSR/RSC rules (Pass 2 Server Actions, Pass 5 Suspense/serialization)
• babel-plugin-react-compiler or react-compiler → skip unnecessary memoization findings
• @radix-ui, shadcn, or similar → flag excessive inline styling more aggressively

Step 2 — Get PR context

Run these commands sequentially (no nested substitution):

1. gh pr view --json number,title,body,baseRefName,headRefName,url — store the baseRefName value
2. git diff <baseRefName>...HEAD — using the baseRefName from step 1
3. git diff --name-only <baseRefName>...HEAD — changed files list

Step 3 — Expand affected scope

For each changed file, identify direct consumers (files that import from it) to catch breakage from interface changes. Only expand one level deep — do not crawl the entire dependency graph.

Step 4 — Run review passes

Apply all 7 review passes and filtering rules from the criteria file against the diff and affected files.

Step 5 — Output findings

Collect all findings with this format per issue:

### [SEVERITY] Pass N — file:line

**Issue:** one-line description
**Why:** reason you're confident this is a real issue
**Fix:** suggested code change

If no issues found: "✅ No issues found — all 7 review passes came back clean."

Show the user the findings.

Step 6 — Autofix

Follow the autofix workflow defined in rules/autofix.md.

• Local mode: present the three options (create findings file as cr-react-findings.md, fix step by step, fix all)

• CI mode: post findings as PR comments with /fix and /fix-all reply instructions

• Do not flag issues that are clearly intentional based on git blame context.

• Run everything autonomously without asking to confirm each step (except the autofix choice in local mode).