playbooks

home / skills / dcjanus / prompts / coderabbit-cli

coderabbit-cli skill

/skills/coderabbit-cli

This skill guides you to use coderabbit CLI for focused code reviews by selecting the right diff scope and prioritizing --prompt-only.

npx playbooks add skill dcjanus/prompts --skill coderabbit-cli

Review the files below or copy the command above to add this skill to your agents.

Files (1)
SKILL.md
2.7 KB
---
name: coderabbit-cli
description: 指导如何使用 CodeRabbit CLI 进行 Code Review。
---

让 AI agent 调用本地 `coderabbit` CLI，对当前改动做代码审查。重点是选对 diff 范围，并优先使用 `--prompt-only`。

## 使用约定

- 这是给 AI agent 用的 review 工具，不是让 agent 代替人类交互式操作 TUI。
- 默认优先 `--prompt-only`，让 CodeRabbit 直接产出给 AI agent 的提示内容。
- 只有在确实需要阅读纯文本 review 原文时，才使用 `--plain`。
- 每次只 review 和当前任务直接相关的 diff，避免把无关改动混进审查范围。
- 单次 review 可能较慢，且存在频率限制；不要在短时间内重复触发多轮无意义审查。
- 当前 CLI 未提供明确的流式输出、进度或超时参数；将 `coderabbit review` 视为长任务处理，不要假设会持续输出日志。
- 运行长时间 review 前，先向用户说明可能需要数分钟到更久；执行期间定期汇报仍在等待结果，避免用户误以为已卡死。
- 如需外层超时控制，使用 shell 的超时机制包裹命令，而不是假设 `coderabbit` 自带 `--timeout`。
- 对用户汇报时，优先总结 CodeRabbit 的有效发现，不要直接大段转储原始输出。
- 向用户汇报整理后的待处理项时，使用数字编号，方便用户按编号指定后续处理项。
- 如果 CodeRabbit 的建议与代码现状或任务目标冲突，agent 应自行判断并说明取舍理由。
- 若最终没有采纳某条建议，应明确写出不采纳原因，例如误报、已有覆盖、与既定约束冲突。
- 推荐流程：先确认当前任务涉及的改动范围，再选择合适的 `review` 参数并运行 `coderabbit review --prompt-only ...`；将输出整理为可执行项，能直接修复的就修复，需要用户决策的再汇报；大改后如有必要，再补一轮 review，默认不要超过 1 到 2 轮。

## 常用命令

```bash
coderabbit review --prompt-only --type uncommitted # 审查当前未提交改动
coderabbit review --prompt-only --base <default-branch> # 审查当前分支相对仓库默认分支的改动
coderabbit review --prompt-only --base-commit HEAD~3 # 审查 HEAD~3 之后引入的改动
coderabbit review --prompt-only --base <default-branch> --config coderabbit.yaml # 基于默认分支审查，并追加仓库内 review 指令文件
coderabbit review --prompt-only --cwd /abs/path/to/repo --base <default-branch> # 指定目标仓库目录，并相对默认分支审查
```

## 参考

- 查看主命令帮助：`coderabbit --help`
- 查看 review 参数：`coderabbit review --help`
- 官方页面：[CodeRabbit CLI page](https://www.coderabbit.ai/cli)

Overview

This skill guides an AI agent to call the local CodeRabbit CLI to perform focused code reviews on the current changes. It emphasizes selecting the correct diff range, preferring the --prompt-only mode, and converting CLI output into actionable items for the user. The goal is to integrate CodeRabbit as a review assistant while keeping the human in the loop.

How this skill works

The agent runs coderabbit review on a specific diff range (uncommitted, relative to default branch, or from a base commit). It defaults to --prompt-only so CodeRabbit produces concise prompts for the agent to parse. The agent treats each review as a long-running task, parses CodeRabbit output into numbered action items, and reports summaries and decisions back to the user.

When to use it

  • Review local uncommitted changes before committing (use --type uncommitted).
  • Audit a branch against the default branch to catch regressions or style drift.
  • Inspect a set of recent commits (e.g., --base-commit HEAD~3) before pushing.
  • When you need actionable suggestions summarized for developer decision-making.
  • When you want CodeRabbit to generate reviewer prompts without large raw dumps.

Best practices

  • Always limit the diff to only the files relevant to the current task to avoid noise.
  • Prefer --prompt-only; use --plain only if you must read the raw review text.
  • Inform the user that the review can take minutes and provide periodic status updates.
  • Wrap long runs with shell-level timeouts if you require external timeout control.
  • Summarize findings as numbered action items; label any non-adopted suggestions with reasons.
  • Avoid repeated quick re-runs; default to at most 1–2 follow-up review rounds after major changes.

Example use cases

  • Run coderabbit review --prompt-only --type uncommitted to check a hotfix before commit.
  • Audit feature branch differences: coderabbit review --prompt-only --base main.
  • Review a sequence of recent commits: coderabbit review --prompt-only --base-commit HEAD~5.
  • Combine repo config: coderabbit review --prompt-only --base main --config coderabbit.yaml to apply repo-specific rules.
  • Run reviews from a specific repo path: coderabbit review --prompt-only --cwd /abs/path/to/repo --base main.

FAQ

Why use --prompt-only by default?

It produces concise agent-friendly prompts so the agent can extract actionable items without dumping long raw text.

What if CodeRabbit suggests changes that conflict with goals?

Evaluate and document the conflict, explain why you did or did not adopt the suggestion, and include that rationale in the report to the user.