Token导航 LogoToken导航

Agent Skill · 研究检索

comprehensive-review(全面审查)

comprehensive-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。

comprehensive-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,权限为需要联网;许可证为 MIT,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 49,058。

总安装 665周安装 28GitHub Stars 6下载 233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
MIT

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:comprehensive-review(全面审查)
来源仓库:https://github.com/troykelly/claude-skills
仓库路径:skills/comprehensive-review
安装命令:
npx skills add https://github.com/troykelly/claude-skills --skill comprehensive-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。不同来源提供的安装方式可能略有差异;本站展示可直接复制的安装命令,安装前请核对来源页面。

skills.shnpx skills
npx skills add https://github.com/troykelly/claude-skills --skill comprehensive-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。

执行七维度代码审查,涵盖功能、安全、性能等多个方面。强制要求将完整评审报告发布至 GitHub issue 方可关闭任务。适用于对交付质量要求严格的协作开发环境。comprehensive-review 属于研究检索类 Skill,可作为该场景下的辅助能力补充。可通过 hook 机制自动拦截未达标 PR 提交。

安装前建议确认 CI/CD 流程是否兼容此审查前置条件。

来源仓库内路径:skills/comprehensive-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/troykelly/claude-skills --skill comprehensive-…;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

    它需要需要联网,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

comprehensive-review 怎么安装?

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

comprehensive-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。

comprehensive-review 需要什么权限?

权限级别是需要联网。该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

comprehensive-review 是免费的吗?

来源标注的许可证是 MIT。本站不代收任何费用,也不托管安装包;能否商用、要不要署名,以许可证原文为准。

comprehensive-review 用的人多吗?

公开数据是累计安装 665 次,近一周安装 28 次,下载 233 次,GitHub 6 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。

comprehensive-review 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。
  • 3 个目录同时收录,说明这个 Skill 流传比较广。
  • 许可证明确为 MIT,商用和分发边界好判断。
  • 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

  • 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。
  • 需要许可证清楚、并且有第三方扫描结果的团队项目。

不太适合

  • 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
comprehensive-review

全面审查 · 本页

6656需要联网MIT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

Comprehensive Review

Overview

Review code against 7 criteria before considering it complete.

Core principle: Self-review catches issues before they reach others.

HARD REQUIREMENT: Review artifact MUST be posted to the GitHub issue. This is enforced by hooks.

Announce at start: "I'm performing a comprehensive code review."

Review Artifact Requirement

This is not optional. Before a PR can be created:

  1. Complete review against all 7 criteria
  2. Document all findings
  3. Post artifact to issue comment using EXACT format below
  4. Address all findings (fix or defer with tracking issues)
  5. Update artifact to show "Unaddressed: 0"

The review-gate skill and PreToolUse hook will BLOCK PR creation without this artifact.

The 7 Criteria

1. Blindspots

Question: What am I missing?

CheckAsk Yourself
Edge casesWhat happens at boundaries? Empty input? Max values?
Error pathsWhat if external services fail? Network issues?
ConcurrencyMultiple users/threads? Race conditions?
StateWhat if called in wrong order? Invalid state?
DependenciesWhat if dependency behavior changes?
// Blindspot example: What if items is empty?
function calculateAverage(items: number[]): number {
  return items.reduce((a, b) => a + b, 0) / items.length;
  // Blindspot: Division by zero when items is empty!
}

// Fixed
function calculateAverage(items: number[]): number {
  if (items.length === 0) {
    throw new Error('Cannot calculate average of empty array');
  }
  return items.reduce((a, b) => a + b, 0) / items.length;
}

2. Clarity/Consistency

Question: Will someone else understand this?

CheckAsk Yourself
NamesDo names describe what things do/are?
StructureIs code organized logically?
ComplexityCan this be simplified?
PatternsDoes this match existing patterns?
SurprisesWould anything surprise a reader?

3. Maintainability

Question: Can this be changed safely?

CheckAsk Yourself
CouplingIs this tightly bound to other code?
CohesionDoes this do one thing well?
DuplicationIs logic repeated anywhere?
TestsDo tests cover this adequately?
ExtensibilityCan new features be added easily?

4. Security Risks

Question: Can this be exploited?

CheckAsk Yourself
Input validationIs all input validated and sanitized?
AuthenticationIs access properly controlled?
AuthorizationAre permissions checked?
Data exposureIs sensitive data protected?
InjectionSQL, XSS, command injection possible?
DependenciesAre dependencies secure and updated?

NOTE: If security-sensitive files are changed (auth, api, middleware, etc.), invoke security-review skill for deeper analysis.

5. Performance Implications

Question: Will this scale?

CheckAsk Yourself
AlgorithmsIs complexity appropriate? O(n²) when O(n) possible?
DatabaseN+1 queries? Missing indexes? Full table scans?
MemoryLarge objects in memory? Memory leaks?
NetworkUnnecessary requests? Large payloads?
CachingShould results be cached?

6. Documentation

Question: Is this documented adequately?

CheckAsk Yourself
Public APIsAre all public functions documented?
ParametersAre parameter types and purposes clear?
ReturnsIs return value documented?
ErrorsAre thrown errors documented?
ExamplesAre complex usages demonstrated?
WhyAre non-obvious decisions explained?

See inline-documentation skill for documentation standards.

7. Standards and Style

Question: Does this follow project conventions?

CheckAsk Yourself
NamingFollows project naming conventions?
FormattingMatches project formatting?
PatternsUses established patterns?
TypesFully typed (no any)?
LanguageUses inclusive language?
IPv6-firstNetwork code uses IPv6 by default? IPv4 only for documented legacy?
LintingPasses all linters?

See style-guide-adherence, strict-typing, inclusive-language, ipv6-first skills.

Review Process

Step 1: Prepare

# Get list of changed files
git diff --name-only HEAD~1

# Get full diff
git diff HEAD~1

# Check for security-sensitive files
git diff --name-only HEAD~1 | grep -E '(auth|security|middleware|api|password|token|secret)'
# If matches found, security-review skill is MANDATORY

Step 2: Review Each Criterion

For each of the 7 criteria:

  1. Review all changed code
  2. Note any issues found
  3. Determine severity (Critical/Major/Minor)

Step 3: Check Security-Sensitive

If ANY security-sensitive files were changed:

  1. Invoke security-review skill OR security-reviewer subagent
  2. Include security review results in artifact
  3. Mark "Security-Sensitive: YES" in artifact

Step 4: Document Findings

## Code Review Findings

### 1. Blindspots
- [ ] **Critical**: No handling for empty array in `calculateAverage()`
- [ ] **Minor**: Missing null check in `formatUser()`

### 2. Clarity/Consistency
- [ ] **Major**: Variable `x` should have descriptive name

### 3. Maintainability
- [x] No issues found

### 4. Security Risks
- [ ] **Critical**: SQL injection possible in `findUser()`

### 5. Performance Implications
- [ ] **Major**: N+1 query in `getOrdersWithUsers()`

### 6. Documentation
- [ ] **Minor**: Missing JSDoc on `processOrder()`

### 7. Standards and Style
- [x] Passes all checks

Step 5: Address All Findings

Use apply-all-findings skill to address every issue.

For findings that cannot be fixed:

  1. Use deferred-finding skill to create tracking issue
  2. Link tracking issue in artifact
  3. "Deferred without tracking issue" is NOT PERMITTED

Step 6: Post Artifact to Issue (MANDATORY)

Post review artifact as comment on the GitHub issue:

ISSUE_NUMBER=123
gh issue comment $ISSUE_NUMBER --body "$(cat <<'EOF'
<!-- REVIEW:START -->
## Code Review Complete

| Property | Value |
|----------|-------|
| Worker | `[WORKER_ID]` |
| Issue | #123 |
| Scope | [MINOR|MAJOR] |
| Security-Sensitive | [YES|NO] |
| Reviewed | [ISO_TIMESTAMP] |

### Criteria Results

| # | Criterion | Status | Findings |
|---|-----------|--------|----------|
| 1 | Blindspots | [✅ PASS|✅ FIXED|⚠️ DEFERRED] | [N] |
| 2 | Clarity | [✅ PASS|✅ FIXED|⚠️ DEFERRED] | [N] |
| 3 | Maintainability | [✅ PASS|✅ FIXED|⚠️ DEFERRED] | [N] |
| 4 | Security | [✅ PASS|✅ FIXED|⚠️ DEFERRED|N/A] | [N] |
| 5 | Performance | [✅ PASS|✅ FIXED|⚠️ DEFERRED] | [N] |
| 6 | Documentation | [✅ PASS|✅ FIXED|⚠️ DEFERRED] | [N] |
| 7 | Style | [✅ PASS|✅ FIXED|⚠️ DEFERRED] | [N] |

### Findings Fixed in This PR

| # | Severity | Finding | Resolution |
|---|----------|---------|------------|
| 1 | [SEVERITY] | [DESCRIPTION] | [HOW_FIXED] |

### Findings Deferred (With Tracking Issues)

| # | Severity | Finding | Tracking Issue | Justification |
|---|----------|---------|----------------|---------------|
| 1 | [SEVERITY] | [DESCRIPTION] | #[ISSUE] | [WHY] |

### Summary

| Category | Count |
|----------|-------|
| Fixed in PR | [N] |
| Deferred (with tracking) | [N] |
| Unaddressed | 0 |

**Review Status:** ✅ COMPLETE
<!-- REVIEW:END -->
EOF
)"

CRITICAL: "Unaddressed" MUST be 0. "Review Status" MUST be "COMPLETE".

Severity Levels

SeverityDescriptionAction
CriticalSecurity issue, data loss, crashMust fix before merge
MajorSignificant bug, performance issueMust fix before merge
MinorStyle, clarity, small improvementShould fix before merge

Checklist

Complete for every code review:

  • Blindspots: Edge cases, errors, concurrency checked
  • Clarity: Names, structure, complexity reviewed
  • Maintainability: Coupling, cohesion, tests evaluated
  • Security: Input, auth, injection, exposure checked (MANDATORY for sensitive files)
  • Performance: Algorithms, queries, memory reviewed
  • Documentation: Public APIs documented
  • Style: Conventions followed
  • All findings documented
  • All findings addressed OR deferred with tracking issues
  • Review artifact posted to issue (exact format)
  • "Unaddressed: 0" in artifact
  • "Review Status: COMPLETE" in artifact

Integration

This skill is called by:

  • issue-driven-development - Step 9

This skill uses:

  • review-scope - Determine review breadth
  • apply-all-findings - Address issues
  • security-review - For security-sensitive changes
  • deferred-finding - For creating tracking issues

This skill is enforced by:

  • review-gate - Verifies artifact before PR
  • PreToolUse hook - Blocks PR without artifact

This skill references:

  • inline-documentation - Documentation standards
  • strict-typing - Type requirements
  • style-guide-adherence - Style requirements
  • inclusive-language - Language requirements
  • ipv6-first - Network code requirements (IPv6 primary, IPv4 legacy)

平台分布

Antigravity

25.88%
按下载量换算60

Claude Code

22.93%
按下载量换算53

Gemini CLI

16.09%
按下载量换算37

Codex

12.69%
按下载量换算30

OpenCode

7.3%
按下载量换算17

Cursor

3.21%
按下载量换算7

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

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。

继续浏览同类 Skills