# PR-Agent — AI PR Reviews via GitHub Action

> PR-Agent is an open-source PR review agent you run as a GitHub Action or CLI to summarize changes and post review suggestions using an LLM key you control.

## Install

Save the content below to `.claude/skills/` or append to your `CLAUDE.md`:

## Quick Use

1. Add a workflow file:
   ```yaml
   name: PR Agent
   on:
     pull_request:
       types: [opened, synchronize]
   jobs:
     pr_agent_job:
       runs-on: ubuntu-latest
       steps:
       - name: PR Agent action step
         uses: the-pr-agent/pr-agent@main
         env:
           OPENAI_KEY: ${{ secrets.OPENAI_KEY }}
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
   ```
2. Add `OPENAI_KEY` as a repo secret.
3. Open a PR and confirm the bot comments within ~30 seconds.

## Intro

PR-Agent is an open-source PR review agent you run as a GitHub Action or CLI to summarize changes and post review suggestions using an LLM key you control.

- **Best for:** teams that want deterministic, repeatable PR reviews (summaries, risk flags, suggestions) as a CI step
- **Works with:** GitHub Actions, pull_request workflows, OPENAI_KEY/LLM key secrets, review comment permissions
- **Setup time:** 10 minutes

## Practical Notes

- Verified adoption signal: uses a single LLM call per tool run (~30 seconds per README claim)
- Docker image namespace migration note: `0.34.2+` uses `pragent/pr-agent` (README)

## How to Run It Safely

PR review bots are only useful when you can trust *when* they run and *what* they can access.

Practical guardrails:

- Keep the workflow limited to `pull_request` (not `pull_request_target`) unless you fully understand the security implications.
- Use GitHub Action permissions narrowly; PR-Agent needs a token that can post comments, not write to the repo.
- Start with one automation (for example: review summary) and add more once the team agrees on the format.

If you prefer local iteration, the README also documents a CLI flow:

```bash
pip install pr-agent
export OPENAI_KEY=your_key_here
pr-agent --pr_url https://github.com/owner/repo/pull/123 review
```

That pattern is useful for testing prompt tweaks before you bake them into CI.

### FAQ

**Q: Is PR-Agent free?**
A: Yes. It’s open-source under an SPDX license; your main cost is the LLM API usage.

**Q: Does it work without GitHub Actions?**
A: Yes. The project also supports a local CLI mode for a single PR URL.

**Q: What’s the minimum permission set?**
A: Typically: read repo contents + write PR comments; avoid broader scopes until needed.

## Source & Thanks

> Source: https://github.com/The-PR-Agent/pr-agent
> License: AGPL-3.0
> GitHub stars: 11,160 · forks: 1,501

---

<!-- ZH -->

## 快速使用

1. 添加 workflow：
   ```yaml
   name: PR Agent
   on:
     pull_request:
       types: [opened, synchronize]
   jobs:
     pr_agent_job:
       runs-on: ubuntu-latest
       steps:
       - name: PR Agent action step
         uses: the-pr-agent/pr-agent@main
         env:
           OPENAI_KEY: ${{ secrets.OPENAI_KEY }}
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
   ```
2. 在仓库 Secrets 里添加 `OPENAI_KEY`。
3. 打开一个 PR，确认机器人在约 30 秒内发出评论。

## 简介

PR-Agent 是开源的 PR 评审 Agent：可用 GitHub Action 或 CLI 在拉取请求上自动生成摘要与建议，核心依赖你自有的 LLM key，适合把“审 PR”变成可复用的流水线。

- **适合谁：** 希望把 PR 评审变成可重复的 CI 步骤（摘要、风险点、建议）的团队
- **可搭配：** GitHub Actions、pull_request 工作流、OPENAI_KEY/LLM key secrets、PR 评论权限
- **准备时间：** 10 分钟

## 实战建议

- 按 README 描述：每个工具通常 1 次 LLM 调用，约 30 秒出结果（供你预估 CI 时延）
- README 提到 Docker 镜像命名空间迁移：`0.34.2+` 使用 `pragent/pr-agent`

## 更安全的落地方式

PR 机器人只有在“可控”时才值得依赖：它什么时候运行、能看到什么、能做什么，都要清晰。

建议的安全护栏：

- 优先用 `pull_request` 触发；除非你非常清楚风险，不要随便用 `pull_request_target`。
- 权限尽量最小化：能评论 PR 即可，不要默认给写仓库权限。
- 先从一种自动化开始（例如 review 摘要），团队接受格式后再逐步扩展。

如果你想先在本地试跑，README 提供了 CLI 流程：

```bash
pip install pr-agent
export OPENAI_KEY=your_key_here
pr-agent --pr_url https://github.com/owner/repo/pull/123 review
```

用它先验证提示词与输出风格，再上 CI 会更稳。

### FAQ

**PR-Agent 免费吗？**
答：是的，它是开源项目（见下方许可证）；主要成本来自你自己配置的 LLM API 调用。

**不使用 GitHub Actions 也能用吗？**
答：可以，README 也提供了本地 CLI 对单个 PR URL 执行的方式。

**最小权限怎么配？**
答：通常只需要读取代码 + 写入 PR 评论；除非必要不要扩展到更高权限。

## 来源与感谢

> Source: https://github.com/The-PR-Agent/pr-agent
> License: AGPL-3.0
> GitHub stars: 11,160 · forks: 1,501


---
Source: https://tokrepo.com/en/workflows/pr-agent-ai-pr-reviews-via-github-action
Author: Agent Toolkit