Token导航 LogoToken导航

Agent Skill · 研究检索

review-code(审查代码)

review-code 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。

review-code 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,权限为接触敏感数据;许可证信息还不明确,第三方扫描已通过,按安装量在研究检索类的 73,675 个 Skill 里排第 54,677。

总安装 501周安装 8GitHub Stars 5下载 65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:review-code(审查代码)
来源仓库:https://github.com/andrewvaughan/agent-council
仓库路径:skills/review-code
安装命令:
npx skills add https://github.com/andrewvaughan/agent-council --skill review-code
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/andrewvaughan/agent-council --skill review-code

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。

用于查找、检索和筛选相关信息,支持快速定位候选结果。适合在关键词、任务场景或来源线索下使用,提升信息获取效率。可结合来源仓库、安装命令和原始 README 核验具体用法。安装前建议确认权限范围、维护状态及是否会触发联网或文件操作。review-code 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

来源仓库内路径:skills/review-code

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/andrewvaughan/agent-council --skill review-cod…;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

    它需要接触敏感数据,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

review-code 怎么安装?

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

review-code 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。

review-code 需要什么权限?

权限级别是接触敏感数据。该 Skill 可能接触密钥、Token、环境变量或敏感配置,应进入高风险复核队列,默认不自动发布。

review-code 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。

review-code 用的人多吗?

公开数据是累计安装 501 次,近一周安装 8 次,下载 65 次,GitHub 5 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。

review-code 安全吗?

第三方扫描结果为通过,但这只覆盖已知的静态检查项,不能替代自己读一遍源码和安装命令。

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。
  • 第三方扫描已通过,常见的高风险行为没有命中。

先想清楚的地方

  • 需要接触敏感数据,装到常用环境前要先评估影响面。
  • 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

  • 环境里有大量密钥、凭证,不方便给第三方 Skill 读取的人。
  • 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
review-code

审查代码 · 本页

5015接触敏感数据未明确通过

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

Code Review Workflow

Run a comprehensive, multi-perspective code review on the current branch changes. This skill activates the Review Council and integrates automated security scanning to catch issues before they reach a pull request.

[!CAUTION] Scope boundary: This skill reviews code and commits fixes. It does NOT create pull requests, push to remote, or merge anything. When the review is complete, stop and suggest the user run /submit-pr next. Do not proceed to PR creation — that is /submit-pr's job.
[!WARNING] Checkpoint protocol. When this workflow reaches a ### CHECKPOINT, you must actively prompt the user for a decision — do not simply present information and continue. Use your agent's interactive prompting mechanism (e.g., AskUserQuestion in Claude Code) to require an explicit response before proceeding. This prevents queued or in-flight messages from being misinterpreted as approval. If your agent lacks interactive prompting, output the checkpoint content and stop all work until the user explicitly responds.

Step 1: Analyze Current Changes

Identify all changes on the current branch:

  1. Run git diff origin/main...HEAD to see all changes relative to main
  2. Run git diff and git diff --cached to catch any uncommitted work
  3. Run git status to see modified, added, and deleted files

Categorize changed files by layer based on your project's directory structure:

  • Frontend: UI components, pages, styles, client-side logic
  • Backend: API routes, services, business logic, server-side code
  • Database: Schema files, migrations, seed data
  • Configuration: Config files, CI/CD, container definitions, environment files
  • Documentation: Markdown files, docs directories
  • Tests: Test files (.test.*, .spec.*)

Present a summary of changed files by category to the user.

Step 2: Run Formatting and Lint Checks

Run your project's formatting and lint commands before the council review to catch mechanical issues early. Common examples:

# Adapt to your project's toolchain
format:check    # Prettier, Black, gofmt, etc.
lint            # ESLint, Ruff, golangci-lint, etc.

If formatting checks fail, fix them immediately by running the auto-formatter on the reported files and stage the changes. Do not include formatting issues in the council review — just fix them.

Step 3: Run Automated Security Scanning

Perform static application security testing (SAST) on all changed files. Scan for:

  • Are all database queries parameterized? (SQL/NoSQL injection)
  • Are user inputs sanitized before rendering? (XSS — reflected, stored, DOM-based)
  • Are state-changing endpoints protected against CSRF?
  • Are there hardcoded secrets, API keys, passwords, or connection strings?
  • Are authentication flows following secure patterns? (no auth bypass paths)
  • Are error messages leaking internal details? (stack traces, DB structure)
  • Are dependencies free of known CVEs?
  • Are file uploads validated and restricted?
  • Are authorization checks present on all protected endpoints?

If backend files are changed, also perform a hardening review:

  • Is input validation comprehensive on all user-facing inputs?
  • Are authentication and authorization checks present and correct?
  • Are error responses safe? (no internal details leaked)
  • Are secrets managed securely? (environment variables, vaults — not hardcoded)
  • Are security headers configured? (CSP, HSTS, X-Frame-Options)
  • Are rate limits configured on public endpoints?

Collect all findings with severity levels (Critical, High, Medium, Low, Info).

Claude Code optimization: If the /security-scanning:security-sast and /security-scanning:security-hardening skills are available, use them for enhanced automated scanning. Otherwise, follow the manual checklists above.

Step 4: Run Accessibility Audit (if frontend changes)

If any frontend files (UI components, CSS, HTML templates) are in the changeset, check WCAG compliance on modified components:

  • Do text and interactive elements meet color contrast ratios? (4.5:1 for normal text, 3:1 for large text)
  • Can all interactive elements be reached and operated via keyboard alone?
  • Are ARIA attributes used correctly? (roles, labels, live regions)
  • Is focus managed properly during dynamic content changes? (modals, route transitions)
  • Is semantic HTML used? (<button> not <div onClick>, <nav>, <main>, <article>)
  • Are images and icons given accessible names? (alt text, aria-label)
  • Are form inputs associated with labels?
  • Does the page make sense when read linearly by a screen reader?
Claude Code optimization: If the /ui-design:accessibility-audit skill is available, use it for enhanced automated WCAG compliance checking. Otherwise, follow the manual checklist above.

Step 5: Activate the Review Council

Read the Review Council template from the skill's councils/review-council.md and conduct the full council review. For each council member, read their agent definition from the skill's agents/ directory and use the complexity tier specified to calibrate review depth.

Security Engineer (Lead)

  • Review all SAST findings from Step 3
  • Check for OWASP Top 10 vulnerabilities in the changed code
  • Validate input sanitization on any new endpoints or forms
  • Check secrets management (no hardcoded credentials, API keys, tokens)
  • Verify authentication and authorization logic
  • Vote: Approve / Concern / Block
  • Findings: List each issue with severity and file location
  • Recommendations: Specific fixes for each finding

QA Lead

  • Assess test coverage for changed files
  • Check if new functionality has corresponding tests
  • Identify untested edge cases and boundary conditions
  • Verify error scenarios are tested
  • Check if test coverage meets the >80% target
  • Vote: Approve / Concern / Block
  • Findings: Missing tests, coverage gaps, untested scenarios
  • Recommendations: Specific tests to add

DevX Engineer

  • Check if documentation is updated for user-facing changes
  • Review code readability and clarity of complex logic
  • Verify README/docs reflect any new features or changes
  • Check that public APIs have proper documentation comments
  • Vote: Approve / Concern / Block
  • Findings: Documentation gaps, unclear code sections
  • Recommendations: Documentation and clarity improvements

Domain Specialist

Select the appropriate specialist based on changed files:

  • Frontend Specialist if frontend files changed
  • Backend Specialist if backend files changed
  • Both if changes span frontend and backend

Review domain-specific best practices:

  • Component patterns, state management, rendering performance (frontend)
  • Framework patterns, service architecture, API design (backend)
  • Schema design, query optimization, migration safety (database)
  • Vote: Approve / Concern / Block
  • Findings: Pattern violations, anti-patterns, performance concerns
  • Recommendations: Specific improvements

Step 6: Present Consolidated Review Report

Present the full review report organized by severity:

Critical / Blocking Issues

Issues that must be fixed before merge (any Block vote or Critical SAST finding).

High Priority Issues

Issues strongly recommended to fix (High SAST findings, Concern votes with security implications).

Medium Priority Issues

Issues worth fixing but not blocking (Medium SAST findings, code quality concerns).

Low Priority / Suggestions

Nice-to-have improvements (Low findings, style suggestions, minor documentation gaps).

Review Decision Summary

  • Overall Status: Approved / Needs Changes / Blocked
  • Council Votes: Summary of each member's vote
  • Action Items: Prioritized list of what to fix

CHECKPOINT: Present all findings to the user. Ask which items they want to address now. Wait for instructions before proceeding.

Step 7: Apply Fixes

For each item the user approves for fixing:

  1. Apply the fix
  2. Re-run the relevant check (lint, test, type-check, or security scan) to verify the fix
  3. Stage the changes

If the user asks to skip certain findings, note them as deferred items for Step 9.

CHECKPOINT: Present the applied fixes to the user. Confirm all changes look correct before committing.

Step 8: Commit

If the user approves:

  1. Stage all changes
  2. Commit with an appropriate conventional commit message (e.g., fix(security): address SAST findings or refactor: address code review feedback)

Step 9: Track Deferred Work

If any review findings were deferred (not fixed in this cycle), document them so they don't get lost.

Check for Deferred Items

Review the following sources for deferred work:

  1. Skipped findings: Items the user chose not to address from the review report
  2. Low priority suggestions: Items categorized as Low/Info that weren't fixed
  3. Council recommendations: Improvements suggested by council members that are out of scope for this PR

Document Deferred Work

For each deferred item, create a tracking issue in your project's issue tracker with the finding description, context on why it was deferred, and what needs to happen.

[!TIP] Not every review generates deferred work. If all findings were addressed or accepted, skip this step entirely. Don't manufacture follow-up issues just to have them.

Step 10: Hand Off — STOP Here

[!CAUTION] This skill's work is done. Do NOT proceed to create a pull request, push to remote, or merge anything. Those are separate skills with their own workflows and checkpoints.

Suggest the next step — then stop:

  • If ready for PR: "Run /submit-pr to create a pull request"
  • If more work needed: "Continue implementation, then run /review-code again when ready"

Do not push the branch, create a PR, or invoke /submit-pr from within this skill.

平台分布

Codex

35.64%
按下载量换算23

Claude

31.91%
按下载量换算21

Cursor

21.47%
按下载量换算14

Gemini CLI

8.96%
按下载量换算6

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

接触敏感数据

该 Skill 可能接触密钥、Token、环境变量或敏感配置,应进入高风险复核队列,默认不自动发布。

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。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