SkillClinic

AI 技能体检诊断，检测Gene结构完整性、触发配置、内容质量

Security Scan

The skill's requests and runtime instructions are coherent with its stated purpose (auditing and optionally fixing other skill files); it asks to read and edit local skill documents but does not request unrelated credentials, installs, or external endpoints.

Purpose & Capability

Name/description: auditing skills (Gene structure, metadata.trigger, content quality). Declared artifacts (SKILL.md + reference docs + report) and allowed-tools (Read/Glob/Write/Edit/AskUserQuestion) align with that purpose. No unrelated binaries, env vars, or external credentials are requested.

Instruction Scope

Runtime instructions say ask the user for a skill name or full path, read SKILL.md and other docs in the directory, compute scores, produce recommendations, and optionally apply edits. Reading and editing skill files is coherent for an audit tool. Caution: because it accepts a full path, it can read/edit arbitrary files the agent has filesystem access to — ensure you only point it at the intended skill directory and review any edits before applying.

Install Mechanism

Instruction-only skill with no install spec and no code files; nothing is downloaded or written by an installer. This is the lowest-risk install model.

Credentials

No environment variables, credentials, or config paths are required. The skill's access (file read/write) is proportional to its purpose of auditing and optionally modifying skill files.

Persistence & Privilege

always:false and user-invocable:true. The skill is not force-included in every agent run. It can edit files (declared in allowed-tools), which is necessary for its 'apply fixes' function; edits should be performed only after explicit user confirmation (the SKILL.md indicates it will ask).

Scan Findings in Context

[no_regex_findings] expected: The static regex scanner had nothing to analyze because this is an instruction-only skill with no code files; absence of findings does not imply safety, but is expected for a docs-only skill.

Assessment

This skill is coherent with its stated purpose but has permissions to read and modify files. Before running it: (1) only provide explicit skill directories you trust — avoid pointing it at system or home directories; (2) back up the target skill files or use version control so you can review/revert edits; (3) inspect any proposed edits the skill shows before confirming 'apply'; (4) be cautious if you intend the agent to run autonomously — while this skill is not always-included, any automation that gives it file-edit permission increases risk. If you want stricter safety, request a dry-run audit (read-only) and refuse the 'apply fixes' step.

技能诊所

Summary

技能质量体检：检测Gene结构（keywords/summary/strategy/AVOID）+ 触发配置（metadata.trigger）+ 内容质量，输出评分和改进建议。

Keywords

技能体检、Gene结构、技能质量、触发配置、metadata.trigger

Strategy

1. 问用户要体检哪个技能：支持技能名称或完整路径
2. 读取技能文件：SKILL.md 和目录下其他文档
3. 检查Gene结构：keywords、summary、strategy、AVOID
4. 检查触发配置：metadata.trigger 是否存在（关键！无此字段技能无法触发）
5. 算分：结构(35) + 触发(10) + 内容(35) + 实践(20) - 负贡献
6. 定等级：S/A/B/C
7. 开处方：提出改进建议，问用户是否执行

AVOID

• AVOID 只读SKILL.md就下结论，目录下其他文档也要看
• AVOID 只改SKILL.md不改其他文档，改进要覆盖所有相关文件
• AVOID 给了分不给建议，等于没说
• AVOID 改文档时表述不清，要说清楚"错误行为 + 应该怎样"
• AVOID 忽略metadata.trigger检查，无此字段技能无法被触发
• AVOID 自己写得烂还去评价别人，先照照镜子

---

入口

问用户：要体检哪个技能？

支持技能名称或完整路径。

标准

维度	满分	检测项
结构分	35	keywords + summary + strategy + AVOID
触发分	10	metadata.trigger（无此字段=技能无法触发！）
内容分	35	Token效率 + 信号密度 + 可执行性
实践分	20	渐进式披露 + Human-in-the-Loop + CLI友好

等级	分数
S	≥80
A	70-79
B	60-69
C	<60

参考

详细标准见 reference/criteria.md