CodeRabbit Skill | Agent Skills

CodeRabbit

AI-powered code review for pull requests and local changes.

Quick Navigation

| Task | Reference | | ----------------------------- | ----------------------------------------------------------- | | Install & run CLI | cli-usage.md | | Configure .coderabbit.yaml | configuration.md | | Supported tools (40+ linters) | tools.md | | Git platform setup | platforms.md | | PR commands (@coderabbitai) | pr-commands.md | | Claude/Cursor/Codex workflow | agent-integration.md | | Triage findings | triage.md | | Fix single issue | fix.md | | Reporting & metrics | end-to-end-workflow.md | | End-to-end workflow | end-to-end-workflow.md | | Windows/WSL setup | windows-wsl.md |

Quick Start

Install

See references/cli-usage.md for installation and authentication.

Run Review

# AI agent workflow (most common)
coderabbit --prompt-only --type uncommitted

# Interactive mode
coderabbit

# Plain text output
coderabbit --plain

Local Capture Script

If you need to persist raw prompt-only output to a file, use the bundled script:

python3 scripts/run_coderabbit.py --output coderabbit-report.txt

Options:

--output to choose a different file name
--timeout to adjust the timeout in seconds (default: 1800)

PR Commands

@coderabbitai review          # Incremental review
@coderabbitai full review     # Complete review
@coderabbitai pause           # Stop auto-reviews
@coderabbitai resume          # Resume auto-reviews
@coderabbitai resolve         # Mark comments resolved

Severity Matrix

| Severity | Action | Examples | | ------------ | --------------- | ------------------------------------------------- | | CRITICAL | Fix immediately | Security, data loss, tenant isolation | | HIGH | Should fix | Reliability, performance, architecture violations | | MEDIUM | Judgment call | Maintainability, type safety (quick wins) | | LOW | Skip | Style/formatting, subjective nits |

AI Agent Workflow Pattern

Implement [feature] and then run the capture script to generate .code-review/coderabbit-report.txt,
run it in a background terminal and wait for the process to complete before reading the report.
Fix any critical issues. Ignore nits.

Key points:

Use --prompt-only for AI-optimized output
Reviews take 7-30+ minutes depending on changeset size
Run command in background terminal (background=true)
Wait for terminal to become idle (not busy) using get_terminal_output
Poll every 60 seconds, not more frequently — CodeRabbit takes time
Do NOT just check for file existence — file is created early but populated gradually
Once terminal shows completion, read .code-review/coderabbit-report.txt
If process times out (30 min default) or errors, report failure to user
Limit to 2-3 review iterations maximum

Linked Repositories (2026-02-18)

CodeRabbit can analyze linked repositories during PR review to catch cross-repo breakages (API/type/dependency drift).

Configure linked repositories in Knowledge Base settings.
Use this when changes in one repo affect contracts in another.
Treat cross-repo findings as HIGH/CRITICAL when they indicate runtime incompatibility.

Minimal Configuration

# .coderabbit.yaml
language: en-US
reviews:
  profile: chill
  high_level_summary: true
  tools:
    gitleaks:
      enabled: true
    ruff:
      enabled: true

Critical Prohibitions

Do not introduce fallbacks, mocks, or stubs in production code
Do not broaden scope beyond what CodeRabbit flagged
Do not "fix" style nits handled by formatters/linters
Do not ignore CRITICAL findings; escalate if unclear
Stop and resolve CLI errors (auth/network) before fixing code

Templates

coderabbit.minimal.yaml — Minimal configuration
coderabbit.full.yaml — Full example with all options
agent-prompts.md — Ready-to-use AI agent prompts

Agent Skills: CodeRabbit

Install this agent skill to your local

Skill Files