← 返回 Skills 市场
ming-shy

Agent-Skill-Reviewer

作者 ming-shy · GitHub ↗ · v1.0.1 · MIT-0
cross-platform ✓ 安全检测通过
148
总下载
0
收藏
0
当前安装
2
版本数
在 OpenClaw 中安装
/install agentskill-reviewer
功能描述
评审 AgentSkills 质量并生成专业报告。用于检查 skill 的内容质量、结构完整性、文档清晰度、冗余问题。当用户要求"评审 skill"、"检查 skill 质量"、"审查 SKILL.md"、"分析这个 skill"时触发。
使用说明 (SKILL.md)

Skill Reviewer - AgentSkill 质量评审系统

本 skill 用于系统化评审 AgentSkill 的质量,核心目标:在保证功能完备的前提下,最大化精简 token 消耗

核心矛盾

  • 过度精简:说明不足,AI 无法正确执行任务
  • 过度冗余:描述繁琐,浪费 context window,逻辑缠绕

目标:系统瘦身 + 逻辑完备 — 让 95% 的人(包括 AI)一眼看懂全流程。

评审流程

1. 初始扫描

读取目标 skill 的 SKILL.md

  • 理解该 skill 的核心功能
  • 识别所有功能入口和工作流
  • 记录 skill 的文件结构(scripts/、references/、assets/)

2. 逐项功能审核

对每个功能或子模块:

可理解性审核

  • 是否可以被 95% 的人理解?
  • 术语是否需要额外解释?
  • 逻辑跳跃是否过大?

Token 效率审核

  • 是否存在重复描述?
  • 是否可以用脚本封装复杂逻辑?(模型只需调用一条命令)
  • 是否可以扁平化层级结构?
  • 是否可以用简洁示例替代冗长说明?

冗余内容审核

  • 是否存在与功能无关的描述?
  • 是否存在重复的文档片段?
  • references/ 文件是否可以进一步精简或合并?

3. SKILL.md 审核

Frontmatter 检查

  • description 是否清晰且包含触发场景?
  • 是否包含所有"何时使用"信息?

Body 检查

  • 是否存在应该移至 references/ 的详细内容?
  • 工作流描述是否简洁有力?
  • 是否存在应删除的冗余段落?

Progressive Disclosure 检查

  • SKILL.md body 是否控制在 500 行以内?
  • 是否正确使用 references/ 分离详细文档?
  • 是否在 SKILL.md 中正确引用外部文件?

4. 逻辑对齐检查(关键)

对比发现的问题原始 skill 功能

  • 禁止为了瘦身而删减原本的任务逻辑
  • 禁止改变原有功能的语义
  • 仅精简表达方式,不改变功能范围

必要时运行 scripts/validate_logic.py 进行自动化逻辑对齐检查。详见 references/validation.md

5. 生成评审报告

使用 scripts/generate_report.py 生成结构化报告,详见 references/report_template.md

评分维度 (10 分制):可理解性、Token 效率、功能完备性、结构合理性

报告结构:结论概览(打分)、优点、存在的问题(按严重程度分级)、优化计划(含 token 节省量和风险评估)

使用示例

# 评审某个 skill
评审 /path/to/my-skill

# 评审并指定输出路径
评审这个 skill 并输出报告到 ./reviews/my-skill-report.md

最佳实践

  1. 先理解再评审 — 不熟悉功能时,先问用户澄清
  2. 保守为主 — 不确定是否可删除时,保留并标注疑问
  3. 量化改进 — 给出具体 token 节省数值
  4. 提供示例 — 展示改进前后对比

输出格式

默认以中文输出,使用 Markdown 格式,清晰分级。报告默认保存到工作目录的 skill-reviews/ 子目录下,文件名为 \x3Cskill-name>-review-\x3Cdate>.md

安全使用建议
This skill appears to do exactly what it claims: parse SKILL.md and related files, run local checks, and generate a Markdown report. Before executing: (1) Inspect the included scripts if you will run them in your environment to confirm they meet your policies (they only perform local file I/O and text analysis). (2) Supply only directories you intend reviewed — do not point it at directories containing secrets you don't want read. (3) Run the scripts in an isolated environment (container or limited-permission account) if you have stricter security requirements. If you need network-isolated auditing, verify nothing in the repo invokes external network calls (the provided files do not).
功能分析
Type: OpenClaw Skill Name: agentskill-reviewer Version: 1.0.1 The 'agent-skill-reviewer' bundle is a legitimate utility designed to analyze and optimize OpenClaw AgentSkills for token efficiency and structural quality. It contains Python scripts (count_tokens.py, generate_report.py, validate_logic.py) that perform local file analysis, token estimation, and logic comparison between original and optimized skill versions. The instructions in SKILL.md are strictly aligned with the stated purpose of quality review and explicitly warn against altering the functional semantics of the skills being reviewed, showing no signs of malicious intent or high-risk vulnerabilities.
能力评估
Purpose & Capability
Name/description match the actual artifacts: SKILL.md describes an AgentSkill reviewer and bundled scripts implement token counting, logic validation, and report generation. No unrelated credentials, binaries, or external services are requested.
Instruction Scope
Instructions and scripts operate on a user-supplied skill directory (reading SKILL.md, references/, scripts/). This is appropriate for a reviewer, but it means the skill will read arbitrary files under the supplied paths — if those target directories contain sensitive secrets, the reviewer code would see them. The SKILL.md and scripts do not call external endpoints.
Install Mechanism
No install spec — instruction-only with Python helper scripts included. No network downloads, package installs, or extraction of remote archives are present.
Credentials
The skill declares no environment variables, credentials, or config paths. The scripts only read/write local files under provided paths and the working directory; this is proportional to the stated purpose.
Persistence & Privilege
Skill is not always-enabled and does not request elevated or persistent system privileges. It writes review output to a local skill-reviews/ directory (documented) which is reasonable for a reporting tool.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install agentskill-reviewer
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /agentskill-reviewer 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.0.1
- Initial release of "agent-skill-reviewer." - Provides a systematic workflow for reviewing AgentSkill quality, focusing on maximizing token efficiency without sacrificing functional completeness. - Introduces a multi-step review process: initial scan, feature auditing, SKILL.md structure check, logic alignment, and automated report generation. - Includes clear scoring dimensions, reporting format, and best practices for reviewers. - Supports Chinese output and structured Markdown reports, with default file locations specified.
v1.0.0
Initial release: full-featured Chinese-language AgentSkills audit & report tool. - Provides a detailed framework for reviewing AgentSkills for content quality, structure, documentation clarity, redundancy, and technical implementation. - Includes comprehensive, step-by-step audit procedures: input validation, file structure checks, structured scoring, and progressive disclosure evaluation. - Generates structured, actionable Chinese Markdown reports with prioritized issue lists, concrete suggestions, and line references. - Identifies common problems such as redundancy, improper file structure, insufficient examples, and improper documentation. - Offers clear usage instructions for reviewing skills and outputting standardized review documents. - Supports multi-file skills (scripts, references, assets) and emphasizes objective, actionable, and constructive feedback.
元数据
Slug agentskill-reviewer
版本 1.0.1
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 2
常见问题

Agent-Skill-Reviewer 是什么?

评审 AgentSkills 质量并生成专业报告。用于检查 skill 的内容质量、结构完整性、文档清晰度、冗余问题。当用户要求"评审 skill"、"检查 skill 质量"、"审查 SKILL.md"、"分析这个 skill"时触发。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 148 次。

如何安装 Agent-Skill-Reviewer?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install agentskill-reviewer」即可一键安装,无需额外配置。

Agent-Skill-Reviewer 是免费的吗?

是的,Agent-Skill-Reviewer 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

Agent-Skill-Reviewer 支持哪些平台?

Agent-Skill-Reviewer 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 Agent-Skill-Reviewer?

由 ming-shy(@ming-shy)开发并维护,当前版本 v1.0.1。

💬 留言讨论