技能评审员
- 作者仓库星标 0
- 作者更新于 2026年8月24日 07:33
- 作者仓库 claude-code-skills
- 领域
- 工程开发
- 作者声明的智能体
- 未找到明确声明;不据此推断已兼容或已测试
- 静态检查
- 88 / 100 · 启发式扫描,不代表运行安全
- 运行环境证据
- 未执行;当前没有隔离运行测试证据
- 作者 / 版本 / 许可证
- @daymade · 未声明许可证
- 上下文用量评级
- 低消耗
- 接入复杂程度
- 即装即用
- 是否需要外部接口密钥
- 不需要
- 兼容的系统
- 未声明(默认跨平台)
- 运行环境要求
- Python
- 检测到的文件与系统行为
-
- 只读
- 允许写入 / 修改
- 网络行为
- 仅限本地
- 安装命令数
- 无(仅作为资料)
需要注意: 未声明工具权限范围;实际权限由运行环境决定。
uv run --with PyYAML python <this-skill-path>/scripts/review_skill.py <target-skill-path>
uv run --with PyYAML python <this-skill-path>/scripts/review_skill.py <target-skill-path> --json
# Skill Reviewer
Review and improve Claude Code skills against official best practices.
## Quick Start
Run the bundled reviewer with PyYAML declared explicitly through `uv`:
```bash
uv run --with PyYAML python <this-skill-path>/scripts/review_skill.py <target-skill-path>
uv run --with PyYAML python <this-skill-path>/scripts/review_skill.py <target-skill-path> --json
```
The reviewer delegates YAML, schema, and internal-path validation to the canonical `skill-creator` validator bundled in the same suite. It then checks frontmatter quality, directory structure, SKILL.md size, hardcoded paths and secrets, script hygiene, `subagent_type` validity, and instruction-style heuristics.
Interpret exit codes as follows: 0 = clean, 1 = warnings only, 2 = review errors, 3 = invocation or runtime failure. Codes 1 and 2 describe the target skill; code 3 means the reviewer could not complete a trustworthy review.
Use the sibling `skill-creator` scripts for the deeper security scan and packaging checks.
## Three Modes
### Mode 1: Self-Review
Check your own skill before publishing.
**Automated review:**
```bash
uv run --with PyYAML python <this-skill-path>/scripts/review_skill.py <target-skill>
```
**Extended security validation:**
```bash
# Security scan
uv run python <this-skill-path>/../skill-creator/scripts/security_scan.py <target-skill> --verbose
```
**Manual evaluation**: See `references/evaluation_checklist.md`.
### Mode 2: External Review
Evaluate someone else's skill repository.
```
Review Workflow:
- [ ] Clone repository to /tmp/
- [ ] Read ALL documentation first
- [ ] Identify author's intent
- [ ] Run evaluation checklist
- [ ] Generate improvement report
```
### Mode 3: Auto-PR
Fork, improve, and submit PR to external skill repository.
```
Auto-PR Workflow:
- [ ] Fork repository (gh repo fork)
- [ ] Create feature branch
- [ ] Apply additive improvements only
- [ ] Self-review: respect check passed?
- [ ] Create PR with detailed explanation
```
## Evaluation Checklist (Quick)
| Category | Check | Status |
|----------|-------|--------|
| **Frontmatter** | name present? | |
| | description present? | |
| | description in third-person? | |
| | includes trigger conditions? | |
| **Instructions** | imperative form? | |
| | under 500 lines? | |
| | workflow pattern? | |
| **Resources** | no hardcoded paths? | |
| | scripts have error handling? | |
Full checklist: `references/evaluation_checklist.md`
## Core Principle: Additive Only
When improving external skills, NEVER:
- Delete existing files
- Remove functionality
- Change primary language
- Rename components
ALWAYS:
- Add new capabilities
- Preserve original content
- Explain every change
```
❌ "Removed metadata.json (non-standard)"
✅ "Added marketplace.json (metadata.json preserved)"
❌ "Rewrote README in English"
✅ "Added README.en.md (Chinese preserved as default)"
```
## Common Issues & Fixes
### Issue: Description Not Third-Person
```yaml
# Before
description: Browse YouTube videos and summarize them.
# After
description: Browses YouTube videos and generates summaries. Use when...
```
### Issue: Missing Trigger Conditions
```yaml
# Before
description: Processes PDF files.
# After
description: Extracts text from PDFs. Use when working with PDF files or when the user mentions PDFs, forms, or document extraction.
```
### Issue: No Workflow Pattern
Add checklist for complex tasks:
```markdown
## Workflow
Copy this checklist:
\`\`\`
Task Progress:
- [ ] Step 1: ...
- [ ] Step 2: ...
\`\`\`
```
### Issue: Missing Marketplace Support
Adding or validating `marketplace.json` (plugin boundaries, `source`/`skills`
layout, whether skills are independently toggleable) is the `marketplace-dev`
skill's domain — don't author it from a template here. Invoke
`daymade-claude-code:marketplace-dev`, then follow its workflow and its cache
and source patterns reference.
## PR Guidelines
When submitting PRs to external repos:
### Tone
```
❌ "Your skill doesn't follow best practices"
✅ "This PR aligns with best practices for better discoverability"
❌ "Fixed the incorrect description"
✅ "Improved description with trigger conditions"
```
### Required Sections
1. **Summary** - What this PR does
2. **What's NOT Changed** - Show respect for original
3. **Rationale** - Why each change helps
4. **Test Plan** - How to verify
Template: `references/pr_template.md`
## Self-Review Checklist
Before submitting any PR:
```
Respect Check:
- [ ] No files deleted?
- [ ] No functionality removed?
- [ ] Original language preserved?
- [ ] Author's design decisions respected?
- [ ] All changes are additive?
- [ ] PR explains the "why"?
```
## References
- `scripts/review_skill.py` - Automated reviewer backed by `skill-creator` validation
- `references/evaluation_checklist.md` - Full evaluation checklist
- `references/pr_template.md` - PR description template
- Best practices: https://platform.claude.com/docs/en/agents-and-tools/agent-skills/best-practices
章节与文件索引
按说明文件中的顺序排列。
快速入门
模式一:自我审查
模式2:外部审查
模式3:自动公关
评估清单(快速)
文件/命令 -> skill-creator · subagent_type · references/evaluation_checklist.md · marketplace.json · source · skills · marketplace-dev · daymade-claude-code:marketplace-dev
内容校验值 -> dbabf60da92c
当前原文的自动摘译,尚未人工审校。
快速入门
使用通过
uv显式声明的 PyYAML 运行捆绑的审核器:审核器将 YAML、模式和内部路径验证委托给捆绑在同一套件中的规范skill-creator验证器。然后,它检查 frontmatter 质量、目录结构、SKILL.md 大小、硬编码路径和秘密、脚本卫生、subagent_type有效性和指令式启发法。按如下方式解释退出代码:0 = 干净、1 = 仅警告、2 = 检查错误、3 = 调用或运行时失败。代码1和2描述了目标技能;代码 3 表示审阅者无法完成值得信赖的审阅。模式一:自我审查
发表之前检查一下自己的技能。自动审查:扩展安全验证:手动评估:参见
references/evaluation_checklist.md。模式2:外部审查
评估别人的技能库。
模式3:自动公关
分叉、改进并将 PR 提交到外部技能存储库。
评估清单(快速)
类别 · 检查 · 状态 Frontmatter · 姓名存在?描述存在吗?第三人称描述?包括触发条件?