Token导航 LogoToken导航

Agent Skill · 研究检索

autonomous-review(自主审查)

autonomous-review 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

autonomous-review 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为操作浏览器;许可证信息还不明确,第三方扫描有提醒项,按安装量在研究检索类的 73,675 个 Skill 里排第 46,660。

总安装 768周安装 33GitHub Stars 15下载 269核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
unknown

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:autonomous-review(自主审查)
来源仓库:https://github.com/zxkane/autonomous-dev-team
仓库路径:skills/autonomous-review
安装命令:
npx skills add https://github.com/zxkane/autonomous-dev-team --skill autonomous-review
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/zxkane/autonomous-dev-team --skill autonomous-review

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

针对自主开发 PR 的自动化代码审查与质量门禁系统。适合需要确保自主生成代码质量的团队协作环境。autonomous-review 属于研究检索类 Skill,可作为该场景下的辅助能力补充。提供跨平台兼容性和浏览器自动化验证支持能力。可选配 IDE 钩子实现自动化的代码风格和质量检查。

使用前应确认是否启用钩子功能及对应的安全策略配置。

来源仓库内路径:skills/autonomous-review

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/zxkane/autonomous-dev-team --skill autonomous-…;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要操作浏览器,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

autonomous-review 怎么安装?

在终端执行 npx skills add https://github.com/zxkane/autonomous-dev-team --skill autonomous-review。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

autonomous-review 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

autonomous-review 需要什么权限?

权限级别是操作浏览器。该 Skill 可能涉及浏览器控制能力,使用时可能读取或操作网页内容,需要在受控环境中确认权限边界。

autonomous-review 是免费的吗?

公开资料没有标注明确许可证,收费和商用限制都无法确认。正式使用前请到来源仓库核对授权说明。

autonomous-review 用的人多吗?

公开数据是累计安装 768 次,近一周安装 33 次,下载 269 次,GitHub 15 星。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

autonomous-review 安全吗?

第三方扫描有提醒项,说明存在需要人工确认的行为。安装前请打开源码看清它读了什么、联网到哪里。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。

先想清楚的地方

  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方扫描存在提醒项或未通过项,需要人工逐条确认。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。

不太适合

  • 需要明确授权才能用到公司项目里的场景。
  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
autonomous-review

自主审查 · 本页

76815操作浏览器未明确有提醒

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

Autonomous Review Mode

You are reviewing a PR created by an autonomous development session. Be thorough and objective.

Cross-Platform Notes

This skill works with any IDE/CLI that supports skills. Browser automation steps use Chrome DevTools MCP — ensure your IDE has this MCP server configured for E2E verification.

Hooks (Optional)

If your IDE supports hooks (Claude Code, Kiro CLI), workflow enforcement hooks in hooks/ provide automatic gate checks. Without hooks, follow each step manually.

Review Checklist

Verify ALL of the following:

1. Process Compliance

  • Design canvas exists in docs/designs/ or docs/plans/
  • Branch follows naming convention (feat/, fix/, refactor/, etc.)
  • Test cases documented in docs/test-cases/
  • PR description follows template (Summary, Design, Test Plan, Checklist sections)
  • PR references the issue (Closes #N or Fixes #N)

2. Code Quality

  • No security issues (no credentials, no injection vulnerabilities)
  • TypeScript types are correct (no any abuse)
  • Error handling is appropriate
  • Code follows existing patterns in the codebase
  • No obvious performance regressions

3. Testing

  • Unit tests exist for new functionality
  • Unit test coverage is reasonable for new code
  • E2E tests updated if UI changes were made
  • All CI checks are passing

4. Infrastructure (if applicable)

  • Infrastructure-as-Code changes are safe
  • No accidental resource deletions
  • IAM permissions follow least privilege

5. Optional: Bot Reviewer Verification

  • If configured bot reviewers have posted reviews, verify their findings are addressed
  • All bot review threads are resolved
  • If bot review is missing and configured, trigger it using scripts/gh-as-user.sh (see below) and wait
IMPORTANT: Some bot reviewers (e.g., Amazon Q Developer) ignore trigger comments posted by GitHub App bot accounts. When triggering bot reviews, you MUST use scripts/gh-as-user.sh so the comment is attributed to a real user: ``bash bash scripts/gh-as-user.sh pr comment {pr_number} --body "/q review" bash scripts/gh-as-user.sh pr comment {pr_number} --body "/codex review" ` Do NOT use the default gh wrapper for bot review triggers — it authenticates as a bot, which some reviewers ignore. If scripts/gh-as-user.sh is not available, fall back to gh pr comment` directly.

6. E2E Verification via Chrome DevTools MCP

If E2E verification is configured (preview URL provided in the prompt), this section is MANDATORY. If no preview URL is configured, skip this section.
  • Preview URL extracted from PR comments
  • Preview URL navigated successfully via Chrome DevTools MCP
  • Test user login verified on preview environment
  • Happy path test cases selected and executed (see section below)
  • Feature test cases executed against live preview
  • Regression tests executed (auth, navigation, console errors)
  • Screenshots captured, uploaded, and linked as evidence
  • E2E verification report posted as PR comment with screenshot links

Merge Conflict Resolution — MANDATORY Pre-Review Step

Before starting the review, check whether the PR branch has merge conflicts with main. If it does, rebase the branch so the PR is mergeable. For the complete rebase procedure, conflict handling, and failure protocol, consult references/merge-conflict-resolution.md.

Quick check:

MERGEABLE=$(gh pr view <PR_NUMBER> --repo <REPO> --json mergeable -q '.mergeable')
  • MERGEABLE — proceed to Review Process
  • CONFLICTING — follow rebase procedure in references
  • UNKNOWN — wait and retry (up to 3 times)

Review Process

  1. Read the issue to understand requirements
  2. Read ALL issue comments to detect requirement changes (see "Requirement Drift Detection" below)
  3. Read the PR diff thoroughly (gh pr diff <number>)
  4. Check CI status (gh pr checks <number>)
  5. Read the files for design docs, test cases, etc. to verify they exist
  6. Assess code quality against the checklist above
  7. Verify bot reviewer findings (if configured — see checklist section 5)
  8. Select happy path test cases based on PR diff analysis (see below)
  9. Perform E2E verification (if configured — see procedure below)
  10. Mark acceptance criteria — for each verified criterion, mark its checkbox in the issue body (see "Marking Acceptance Criteria")
  11. MANDATORY SELF-CHECK GATE — execute the Findings->Decision Gate (see below) BEFORE submitting any review verdict

Requirement Drift Detection — MANDATORY

This step MUST be performed BEFORE reading the PR diff. Requirements can change after implementation via issue comments from the repo owner or maintainers.

Read ALL comments on the issue (not just the body) and look for:

  • Scope changes ("remove", "no longer", "drop", "don't support", "instead of")
  • New requirements added after the original issue was created
  • Corrections or clarifications from the repo owner
  • Explicit instructions to the dev agent that may not yet be reflected in the PR code
# Read all issue comments to check for requirement changes
gh issue view <ISSUE_NUMBER> --repo <REPO> --json comments \
  -q '.comments[] | "\(.author.login) [\(.createdAt)]: \(.body[0:500])"'

If any requirement change is found that the PR code does NOT reflect:

  • This is a [BLOCKING] Requirement drift finding
  • The PR must be sent back to dev with specific instructions about what changed
  • Quote the comment that changed the requirement
  • List the specific code/files that need to be updated

Happy Path Test Cases

Happy path test cases are project-specific. The review agent selects cases based on:

  1. Read docs/test-cases/ directory for available test case documents
  2. Analyze the PR diff to determine which areas changed
  3. Select the most relevant test cases covering changed functionality
  4. Execute at least one happy path test case per review

If no test case documents exist, execute a basic smoke test:

  • Navigate to the application root URL
  • Verify the page loads without errors
  • Check browser console for JavaScript errors

E2E Verification Procedure

This section applies only when E2E verification is configured. The review wrapper script (autonomous-review.sh) will indicate whether E2E is enabled and provide the necessary configuration in the prompt.

For the complete step-by-step E2E procedure (browser automation, screenshot upload, test execution, report format), consult references/e2e-verification.md.

Key steps:

  1. Verify preview URL is available
  2. Open browser and navigate via Chrome DevTools MCP
  3. Login with test user credentials
  4. Execute happy path and feature test cases
  5. Run regression checks (auth, navigation, console errors)
  6. Post structured E2E report on the PR with screenshot evidence

Marking Acceptance Criteria

During E2E verification, mark each acceptance criterion checkbox in the issue body as you verify it.

Procedure

  1. Read the issue body and identify the ## Acceptance Criteria section
  2. For each criterion: a. Verify it via Chrome DevTools MCP, code inspection, or CI check results b. If it passes, mark the checkbox: bash scripts/mark-issue-checkbox.sh <ISSUE_NUMBER> "<criterion text>" c. If it fails, STOP marking — record the failure and proceed to "Review findings"
  3. The script uses gh (which picks up the active App token via GH_TOKEN_FILE), so edits appear as the configured review bot

Important Rules

  • Mark criteria only after verifying them — do not pre-mark
  • If ANY criterion fails, do NOT mark it — post "Review findings:" instead
  • Do NOT mark Requirements checkboxes — those are for the dev agent
  • ALL acceptance criteria must be checked (- [x]) before approving the PR

Findings -> Decision Gate — MANDATORY

This gate is NON-NEGOTIABLE. Execute this self-check BEFORE submitting any PR review (APPROVE or REQUEST_CHANGES) and BEFORE posting the verdict comment on the issue.

For the complete gate procedure (finding classification, blocking vs non-blocking rules, self-check questions, decision criteria, and output format), consult references/decision-gate.md.

Summary of the hard rule:

  • ANY blocking finding -> verdict MUST be FAIL (do NOT approve)
  • ZERO blocking findings -> verdict is PASS (approve + merge)
  • There is NO middle ground — blocking findings and APPROVE are mutually exclusive

Post the review result as a comment on the issue (NOT the PR). Use "Review PASSED" for pass, "Review findings:" for fail.


References

For detailed procedures, consult:

  • references/merge-conflict-resolution.md -- Complete rebase procedure, conflict handling, and failure protocol
  • references/e2e-verification.md -- Browser automation steps, screenshot upload, test execution, E2E report format
  • references/decision-gate.md -- Finding classification, blocking rules, decision criteria, and output format

平台分布

Codex

35.88%
按下载量换算97

Claude

28.62%
按下载量换算77

Cursor

18.04%
按下载量换算49

Gemini CLI

9.2%
按下载量换算25

安全审计

Gen Agent Trust Hub

通过

Socket

可疑

Snyk

可疑

权限和风险

操作浏览器

该 Skill 可能涉及浏览器控制能力,使用时可能读取或操作网页内容,需要在受控环境中确认权限边界。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。来源安全扫描存在提醒或未通过项,安装前请自行复核。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills