skill-review
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
ChineseSkill Review Skill
Skill Review 技能
Overview
概述
The skill-review skill provides a comprehensive, systematic process for auditing skills in the claude-skills repository. It combines automated technical validation with AI-powered verification to ensure skills remain accurate, current, and high-quality.
Use this skill when:
- Investigating suspected issues in a skill
- Major package version updates released (e.g., better-auth 1.x → 2.x)
- Skill last verified >90 days ago
- Before submitting skill to marketplace
- User reports errors following skill instructions
- Examples seem outdated or contradictory
Production evidence: Successfully audited better-auth skill (2025-11-08), found 6 critical/high issues including non-existent API imports, removed 665 lines of incorrect code, implemented v2.0.0 with correct patterns.
s skill-review skill为审核claude-skills仓库中的skill提供了一套全面、系统的流程。它结合了自动化技术校验和AI驱动的验证能力,确保skill始终准确、最新且高质量。
适用场景:
- 排查skill中疑似存在的问题
- 重大包版本发布更新(例如better-auth 1.x → 2.x)
- skill上次验证时间超过90天
- 向市场提交skill之前
- 用户反馈按照skill指引操作出现错误
- 示例看起来过时或存在矛盾
生产验证案例:已成功审计better-auth skill(2025-11-08),发现6个严重/高优问题,包括不存在的API导入,移除了665行错误代码,按照正确模式实现了v2.0.0版本。
Quick Start
快速开始
Invoke via Slash Command
通过斜杠命令调用
/review-skill <skill-name>Example:
/review-skill better-auth/review-skill <skill-name>示例:
/review-skill better-authInvoke via Skill (Proactive)
通过Skill主动触发
When Claude notices potential issues, it can suggest:
User: "I'm having trouble with better-auth and D1"
Claude: "I notice the better-auth skill was last verified 6 months ago.
Would you like me to review it? Better-auth recently released v1.3
with D1 changes."当Claude发现潜在问题时,会主动建议:
用户:"我在使用better-auth和D1的时候遇到了问题"
Claude:"我注意到better-auth skill上次验证是6个月前。better-auth最近发布了v1.3版本,对D1做了改动。你需要我审核这个skill吗?"What This Skill Does
该技能的功能
9-Phase Systematic Audit
9阶段系统审计
-
Pre-Review Setup (5-10 min)
- Install skill locally:
./scripts/install-skill.sh <skill-name> - Check current version and last verified date
- Test skill discovery
- Install skill locally:
-
Standards Compliance (10-15 min)
- Validate YAML frontmatter (name, description, license)
- Check keyword comprehensiveness
- Verify third-person description style
- Ensure directory structure matches spec
-
Official Documentation Verification (15-30 min)
- Use Context7 MCP or WebFetch to verify API patterns
- Check GitHub for recent updates and issues
- Verify package versions against npm registry
- Compare with production repositories
-
Code Examples & Templates Audit (20-40 min)
- Verify import statements exist in current packages
- Check API method signatures match official docs
- Ensure schema consistency across files
- Test templates build and run
-
Cross-File Consistency (15-25 min)
- Compare SKILL.md vs README.md examples
- Verify "Bundled Resources" section matches actual files
- Ensure configuration examples consistent
-
Dependencies & Versions (10-15 min)
- Run
./scripts/check-versions.sh <skill-name> - Check for breaking changes in package updates
- Verify "Last Verified" date is recent
- Run
-
Issue Categorization (10-20 min)
- Classify by severity: 🔴 Critical / 🟡 High / 🟠 Medium / 🟢 Low
- Document with evidence (GitHub URL, docs link, npm changelog)
-
Fix Implementation (30 min - 4 hours)
- Auto-fix unambiguous issues
- Ask user only for architectural decisions
- Update all affected files consistently
- Bump version if breaking changes
-
Post-Fix Verification (10-15 min)
- Test skill discovery
- Verify templates work
- Check no contradictions remain
- Commit with detailed changelog
-
审核前准备 (5-10分钟)
- 本地安装skill:
./scripts/install-skill.sh <skill-name> - 检查当前版本和上次验证日期
- 测试skill发现能力
- 本地安装skill:
-
标准合规性检查 (10-15分钟)
- 校验YAML前言(名称、描述、许可证)
- 检查关键词完整性
- 验证第三人称描述风格
- 确保目录结构符合规范
-
官方文档验证 (15-30分钟)
- 使用Context7 MCP或WebFetch验证API模式
- 查看GitHub上的近期更新和issue
- 对照npm registry验证包版本
- 与生产仓库进行比对
-
代码示例与模板审计 (20-40分钟)
- 验证导入语句在当前包中真实存在
- 检查API方法签名与官方文档一致
- 确保跨文件的schema一致性
- 测试模板可以正常构建和运行
-
跨文件一致性检查 (15-25分钟)
- 比对SKILL.md和README.md中的示例
- 验证「捆绑资源」部分与实际文件匹配
- 确保配置示例一致
-
依赖与版本检查 (10-15分钟)
- 执行
./scripts/check-versions.sh <skill-name> - 检查包更新中的破坏性变更
- 验证「上次验证」日期为近期
- 执行
-
问题分类 (10-20分钟)
- 按严重性分类:🔴 严重 / 🟡 高优 / 🟠 中优 / 🟢 低优
- 留存证据记录(GitHub URL、文档链接、npm更新日志)
-
修复实现 (30分钟 - 4小时)
- 自动修复无歧义的问题
- 仅在需要架构决策时询问用户
- 统一更新所有受影响的文件
- 如果存在破坏性变更则升级版本号
-
修复后验证 (10-15分钟)
- 测试skill发现能力
- 验证模板正常工作
- 检查不存在遗留矛盾
- 提交包含详细变更日志的commit
Automated Checks (via script)
自动化检查(通过脚本)
The skill runs which checks:
./scripts/review-skill.sh <skill-name>- ✅ YAML frontmatter syntax and required fields
- ✅ Package version currency (npm)
- ✅ Broken links (HTTP status)
- ✅ TODO markers in code
- ✅ File organization (expected directories exist)
- ✅ "Last Verified" date staleness
该技能会执行,检查以下内容:
./scripts/review-skill.sh <skill-name>- ✅ YAML前言语法和必填字段
- ✅ 包版本时效性(npm)
- ✅ 损坏链接(HTTP状态)
- ✅ 代码中的TODO标记
- ✅ 文件组织(预期目录存在)
- ✅ 「上次验证」日期是否过期
Manual Verification (AI-powered)
人工验证(AI驱动)
Claude performs:
- 🔍 API method verification against official docs
- 🔍 GitHub activity and issue checks
- 🔍 Production repository comparisons
- 🔍 Code example correctness
- 🔍 Schema consistency validation
Claude会执行:
- 🔍 对照官方文档验证API方法
- 🔍 检查GitHub活动和issue
- 🔍 生产仓库比对
- 🔍 代码示例正确性
- 🔍 Schema一致性校验
Process Workflow
流程工作流
Step 1: Run Automated Checks
步骤1:运行自动化检查
bash
./scripts/review-skill.sh <skill-name>Interpret output to identify technical issues.
bash
./scripts/review-skill.sh <skill-name>解读输出结果,识别技术问题。
Step 2: Execute Manual Verification
步骤2:执行人工验证
For Phase 3: Official Documentation Verification:
-
Use Context7 MCP (if available):
Use Context7 to fetch: /websites/<package-docs> Search for: [API method from skill] -
Or use WebFetch:
Fetch: https://<official-docs-url> Verify: [specific patterns] -
Check GitHub:
Visit: https://github.com/<org>/<repo>/commits/main Check: Last commit, recent changes Search issues: [keywords from skill] -
Find production examples:
WebSearch: "<package> cloudflare production github" Compare: Do real projects match our patterns?
For Phase 4: Code Examples Audit:
- Verify all imports exist (check official docs)
- Check API method signatures match
- Ensure schema consistency across files
- Test templates actually work
针对阶段3:官方文档验证:
-
优先使用Context7 MCP(如果可用):
使用Context7获取:/websites/<package-docs> 搜索:[skill中的API方法] -
或使用WebFetch:
获取:https://<official-docs-url> 验证:[特定模式] -
检查GitHub:
访问:https://github.com/<org>/<repo>/commits/main 检查:最近一次commit、近期变更 搜索issue:[skill中的关键词] -
查找生产示例:
网页搜索:"<package> cloudflare production github" 比对:真实项目是否与我们的模式匹配?
针对阶段4:代码示例审计:
- 验证所有导入真实存在(检查官方文档)
- 检查API方法签名匹配
- 确保跨文件的schema一致性
- 测试模板确实可以正常工作
Step 3: Categorize Issues
步骤3:问题分类
🔴 CRITICAL - Breaks functionality:
- Non-existent API methods/imports
- Invalid configuration
- Missing required dependencies
🟡 HIGH - Causes confusion:
- Contradictory examples across files
- Inconsistent patterns
- Outdated major versions
🟠 MEDIUM - Reduces quality:
- Stale minor versions (>90 days)
- Missing documentation sections
- Incomplete error lists
🟢 LOW - Polish issues:
- Typos, formatting inconsistencies
- Missing optional metadata
🔴 严重 - 破坏功能:
- 不存在的API方法/导入
- 无效配置
- 缺少必填依赖
🟡 高优 - 造成混淆:
- 跨文件的示例矛盾
- 不一致的模式
- 过时的主版本
🟠 中优 - 降低质量:
- 过期的次版本(>90天)
- 缺失文档章节
- 不完整的错误列表
🟢 低优 - 优化问题:
- 拼写错误、格式不一致
- 缺失可选元数据
Step 4: Fix Issues
步骤4:修复问题
Auto-fix when:
- ✅ Fix is unambiguous (correct import from docs)
- ✅ Evidence is clear
- ✅ No architectural impact
Ask user when:
- ❓ Multiple valid approaches
- ❓ Breaking change decision
- ❓ Architectural choice
Format for questions:
I found [issue]. There are [N] approaches:
1. [Approach A] - [Pros/Cons]
2. [Approach B] - [Pros/Cons]
Recommendation: [Default based on evidence]
Which would you prefer?自动修复适用于:
- ✅ 修复方案无歧义(文档中明确的正确导入)
- ✅ 证据清晰
- ✅ 无架构层面影响
询问用户适用于:
- ❓ 存在多个有效方案
- ❓ 需要决定是否接受破坏性变更
- ❓ 涉及架构选择
问题询问格式:
我发现了[问题]。有[N]种解决方案:
1. [方案A] - [优缺点]
2. [方案B] - [优缺点]
建议:[基于证据的默认方案]
你更倾向于哪一种?Step 5: Version Bump Assessment
步骤5:版本升级评估
If breaking changes:
- Major: v1.0.0 → v2.0.0 (API patterns change)
- Minor: v1.0.0 → v1.1.0 (new features, backward compatible)
- Patch: v1.0.0 → v1.0.1 (bug fixes only)
如果存在破坏性变更:
- 主版本:v1.0.0 → v2.0.0(API模式变更)
- 次版本:v1.0.0 → v1.1.0(新增功能,向后兼容)
- 补丁版本:v1.0.0 → v1.0.1(仅bug修复)
Step 6: Generate Audit Report
步骤6:生成审计报告
markdown
undefinedmarkdown
undefinedSkill Review Report: <skill-name>
Skill审核报告:<skill-name>
Date: YYYY-MM-DD
Trigger: [Why review performed]
Time Spent: [Duration]
日期:YYYY-MM-DD
触发原因:[执行审核的原因]
耗时:[时长]
Findings
发现问题
🔴 CRITICAL (N): [List with evidence]
🟡 HIGH (N): [List with evidence]
🟠 MEDIUM (N): [List with evidence]
🟢 LOW (N): [List with evidence]
🔴 严重(N个):[带证据的问题列表]
🟡 高优(N个):[带证据的问题列表]
🟠 中优(N个):[带证据的问题列表]
🟢 低优(N个):[带证据的问题列表]
Remediation
修复方案
Files Modified: [List]
Version Update: [old] → [new]
Breaking Changes: Yes/No
修改的文件:[列表]
版本更新:[旧版本] → [新版本]
破坏性变更:是/否
Verification
验证结果
✅ Discovery test passed
✅ Templates work
✅ Committed: [hash]
✅ 发现测试通过
✅ 模板正常工作
✅ 已提交:[commit hash]
Recommendation
建议
[Final assessment]
---[最终评估]
---Example: better-auth Audit
示例:better-auth审计
Findings
发现问题
Issue #1: Non-existent d1Adapter 🔴 CRITICAL
Location:
references/cloudflare-worker-example.ts:17Problem: Imports from which doesn't exist
d1Adapter'better-auth/adapters/d1'Evidence:
- Official docs: https://better-auth.com/docs/integrations/drizzle
- GitHub: No export in codebase
d1Adapter - Production: 4 repos use Drizzle/Kysely
Fix: Replace with from
drizzleAdapter'better-auth/adapters/drizzle'问题#1:不存在的d1Adapter 🔴 严重
位置:
references/cloudflare-worker-example.ts:17问题:从导入,但该导入不存在
'better-auth/adapters/d1'd1Adapter证据:
- 官方文档:https://better-auth.com/docs/integrations/drizzle
- GitHub:代码库中没有导出
d1Adapter - 生产环境:4个仓库使用Drizzle/Kysely
修复方案:替换为从导入的
'better-auth/adapters/drizzle'drizzleAdapterResult
结果
- Files deleted: 3 (obsolete patterns)
- Files created: 3 (correct patterns)
- Lines changed: +1,266 net
- Version: v1.0.0 → v2.0.0
- Time: 3.5 hours
- 删除的文件:3个(过时模式)
- 新建的文件:3个(正确模式)
- 变更行数:净增加1266行
- 版本:v1.0.0 → v2.0.0
- 耗时:3.5小时
Bundled Resources
捆绑资源
This skill references:
- - Complete 9-phase manual guide
planning/SKILL_REVIEW_PROCESS.md - - Automated validation script
scripts/review-skill.sh - - Slash command definition
.claude/commands/review-skill.md
该技能引用以下资源:
- - 完整的9阶段人工指南
planning/SKILL_REVIEW_PROCESS.md - - 自动化校验脚本
scripts/review-skill.sh - - 斜杠命令定义
.claude/commands/review-skill.md
When Claude Should Invoke This Skill
Claude触发该技能的场景
Proactive triggers:
- User mentions skill seems outdated
- Package major version mentioned
- User reports errors following skill
- Checking metadata shows >90 days since verification
Explicit triggers:
- "review the X skill"
- "audit better-auth skill"
- "is cloudflare-worker-base up to date?"
- "check if tailwind-v4-shadcn needs updating"
主动触发:
- 用户提到skill看起来过时
- 提到包的主版本更新
- 用户反馈按照skill操作出现错误
- 检查元数据发现距离上次验证超过90天
显式触发:
- "review the X skill"
- "audit better-auth skill"
- "cloudflare-worker-base是否是最新的?"
- "检查tailwind-v4-shadcn是否需要更新"
Token Efficiency
Token效率
Without this skill: ~25,000 tokens
- Trial-and-error verification
- Repeated doc lookups
- Inconsistent fixes across files
- Missing evidence citations
With this skill: ~5,000 tokens
- Systematic process
- Clear decision trees
- Evidence-based fixes
- Comprehensive audit trail
Savings: ~80% (20,000 tokens)
不使用该技能:约25,000 tokens
- 反复试错验证
- 重复的文档查询
- 跨文件的修复不一致
- 缺失证据引用
使用该技能:约5,000 tokens
- 系统化流程
- 清晰的决策树
- 基于证据的修复
- 全面的审计轨迹
节省:约80%(20,000 tokens)
Common Issues Prevented
可预防的常见问题
- Fake API adapters - Non-existent imports
- Stale API methods - Changed signatures
- Schema inconsistency - Different table names
- Outdated scripts - Deprecated approaches
- Version drift - Packages >90 days old
- Contradictory examples - Multiple conflicting patterns
- Broken links - 404 documentation URLs
- YAML errors - Invalid frontmatter syntax
- Missing keywords - Poor discoverability
- Incomplete bundled resources - Listed files don't exist
- 伪造的API适配器 - 不存在的导入
- 过时的API方法 - 变更的签名
- Schema不一致 - 不同的表名
- 过时的脚本 - 废弃的实现方式
- 版本漂移 - 包超过90天未更新
- 矛盾的示例 - 多个冲突的实现模式
- 损坏的链接 - 404的文档URL
- YAML错误 - 无效的前言语法
- 缺失关键词 - 可发现性差
- 不完整的捆绑资源 - 列出的文件不存在
Best Practices
最佳实践
- Always cite sources - GitHub URL, docs link, npm changelog
- No assumptions - Verify against current official docs
- Be systematic - Follow all 9 phases
- Fix consistency - Update all files, not just one
- Document thoroughly - Detailed commit messages
- Test after fixes - Verify skill still works
- 始终引用来源 - GitHub URL、文档链接、npm更新日志
- 不做假设 - 对照当前官方文档验证
- 系统化执行 - 遵循全部9个阶段
- 保持修复一致性 - 更新所有相关文件,而非仅修改一处
- 完整记录 - 详细的commit信息
- 修复后测试 - 验证skill仍可正常工作
Known Limitations
已知限制
- Link checking requires network access
- Package version checks need npm installed
- Context7 MCP availability varies by package
- Production repo search may need GitHub API
- Manual phases require human judgment
- 链接检查需要网络访问
- 包版本检查需要安装npm
- Context7 MCP的可用性因包而异
- 生产仓库搜索可能需要GitHub API
- 人工阶段需要人为判断
Version History
版本历史
v1.0.0 (2025-11-08)
- Initial release
- 9-phase systematic audit process
- Automated script + manual guide
- Slash command + skill wrapper
- Production-tested on better-auth v2.0.0 audit
v1.0.0 (2025-11-08)
- 首次发布
- 9阶段系统审计流程
- 自动化脚本 + 人工指南
- 斜杠命令 + skill封装
- 在better-auth v2.0.0审计中经过生产验证
Additional Resources
额外资源
- Full Process Guide:
planning/SKILL_REVIEW_PROCESS.md - Repository: https://github.com/jezweb/claude-skills
- Example Audit: See process guide Appendix B (better-auth v2.0.0)
Last verified: 2025-11-08 | Version: 1.0.0
- 完整流程指南:
planning/SKILL_REVIEW_PROCESS.md - 仓库:https://github.com/jezweb/claude-skills
- 审计示例:见流程指南附录B(better-auth v2.0.0)
上次验证:2025-11-08 | 版本:1.0.0