coderabbit

📁 itechmeat/llm-code 📅 Jan 26, 2026
28
总安装量
28
周安装量
#7288
全站排名
安装命令
npx skills add https://github.com/itechmeat/llm-code --skill coderabbit

Agent 安装分布

github-copilot 25
opencode 8
claude-code 7
gemini-cli 6
cursor 5
antigravity 4

Skill 文档

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#reporting–metrics
End-to-end workflow end-to-end-workflow.md
Windows/WSL setup windows-wsl.md

Quick Start

Install

curl -fsSL https://cli.coderabbit.ai/install.sh | sh
source ~/.zshrc
coderabbit auth login

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

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

Links

Templates