Coderabbit

Skill lgtm-hq/ai-skills/skills/coderabbit

Canonical Agent Skills library for Claude Code, Cursor, Codex, and other agents.

Install

npx -y skills add lgtm-hq/ai-skills --skill coderabbit

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Run CodeRabbit CLI for pre-push AI diff review. Use when asked for CodeRabbit, cr review, or as part of the default dual pre-push workflow with greptile. Max 2-3 runs per change set.

SKILL.md

2.9 KB, as published. Nobody here has run it

CodeRabbit CLI Review

Run CodeRabbit (cr / coderabbit) for automated diff review before push. Pair with greptile for the default pre-CI workflow — both tools catch different issues and mirror CI.

Relationship to analyze-* skills

The analyze-code, analyze-project, and analyze-tests skills are for manual or agent-driven pre-review checks—they run concrete commands (lintro, ripgrep, coverage, etc.) and produce structured findings before you open a PR.

This skill uses CodeRabbit CLI for external AI diff review. Follow the lint and test skills (or the commit skill, which includes them) before invoking CodeRabbit. Do not duplicate full analyze rubrics here.

Relationship to greptile

Default pre-push flow when CI runs both tools:

commit → [greptile ‖ coderabbit] → pr

Each step names the skill to follow.

Run greptile and coderabbit in parallel when possible. Do not re-run on unchanged code.

Commands

cr -h / coderabbit --help — Show CodeRabbit help
coderabbit review --agent --type committed --base main — Full branch diff against main (default for agents; matches CI intent)
coderabbit review --agent --type uncommitted — Uncommitted (staged + unstaged); quick spot check
coderabbit doctor — Verify install, auth, and Git repo readiness
coderabbit review findings — Replay findings from the last local review

cr is an alias for coderabbit. --prompt-only is deprecated; use --agent.

Rules

Max 2–3 CodeRabbit runs per change set (Pro plan: 5 CLI reviews per developer — rate limits)
Always use --agent for agent workflows (structured JSON output)
Default to --base main unless the repo uses a different default branch
Reviews can take 7–30+ minutes — run in the background when possible
Must run from an initialized Git repository root
Only use -t uncommitted for quick pre-commit spot checks

Usage

When asked to run CodeRabbit or as part of pre-push review:

Ensure lint and tests pass (the commit skill's checklist, or the lint and test skills explicitly)
Run coderabbit review --agent --type committed --base main (background if needed)
Analyze findings; address critical/major issues
After fixes, run one verification pass if CodeRabbit had findings
Track run count (max 2–3 per change set)

Note: -t committed with --base main reviews the full branch diff. -t uncommitted includes staged and unstaged changes (empty after a full commit with a clean tree).