| name | coderabbit |
| description | Local AI code reviews via CodeRabbit CLI. ONLY use when (1) user explicitly requests "coderabbit"/"cr review", OR (2) code changes are high-risk (security, concurrency, complex logic). Rate-limited to 1 review/hour—be highly selective. |
CodeRabbit CLI
AI code reviews locally. Rate limit: 1 review/hour. Use sparingly.
When to Use (Be Selective)
USE CodeRabbit for:
- User explicitly requests it
- Security-sensitive code (auth, crypto, input validation, secrets)
- Concurrency/async (race conditions, deadlocks, shared state)
- Memory management (leaks, resource cleanup)
- Complex business logic with non-obvious edge cases
- Database migrations or schema changes
- Public API contract changes
SKIP CodeRabbit for:
- Simple refactors, renames, moves
- Documentation or comment changes
- Test additions (unless testing complex logic)
- Config/env changes
- Obvious bug fixes
- Formatting/style cleanup
When uncertain, ask user: "This looks like it could benefit from a CodeRabbit review (1/hour limit). Want me to run one?"
Setup (First Use)
# Check if installed
command -v cr &>/dev/null || {
echo "Installing CodeRabbit CLI..."
curl -fsSL https://cli.coderabbit.ai/install.sh | sh
source ~/.zshrc # or ~/.bashrc
}
# Check auth status, login if needed
cr auth status || cr auth login
If CLI missing, prompt user: "CodeRabbit CLI not installed. Install it now?" Then run install + auth.
Execution (Always Background)
ALWAYS run in background with monitoring:
# Start review in background, capture output
cr review --prompt-only --type uncommitted > /tmp/cr-review.txt 2>&1 &
CR_PID=$!
# Monitor completion (reviews take 7-30 min)
while kill -0 $CR_PID 2>/dev/null; do sleep 30; done
# Read results
cat /tmp/cr-review.txt
Flags:
--prompt-only— Always use. AI-optimized, token-efficient output--type uncommitted— Faster, reviews only unstaged/staged changes--type committed— Only committed changes vs base--base <branch>— Compare against specific branch (default: main)
Workflow
- Run background review (command above)
- Continue other work while waiting
- Parse findings when complete — prioritize: critical > major > minor (ignore nits)
- Fix critical/major only — don't chase diminishing returns
- NO verification re-run unless user requests (preserve rate limit)
Troubleshooting
- No findings: Check
git status, verify--basebranch - Stalled: Reviews can take 30+ min for large diffs — be patient
- Auth issues: Re-run
cr auth login