Agent-Skill-Reviewer

v1.0.1

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。

0· 140·0 current·0 all-time

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for ming-shy/agentskill-reviewer.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "Agent-Skill-Reviewer" (ming-shy/agentskill-reviewer) from ClawHub.
Skill page: https://clawhub.ai/ming-shy/agentskill-reviewer
Keep the work scoped to this skill only.
After install, inspect the skill metadata and help me finish setup.
Use only the metadata you can verify from ClawHub; do not invent missing requirements.
Ask before making any broader environment changes.

Command Line

CLI Commands

Use the direct CLI path if you want to install manually and keep every step visible.

OpenClaw CLI

Bare skill slug

openclaw skills install agentskill-reviewer

ClawHub CLI

Package manager switcher

npx clawhub@latest install agentskill-reviewer
Security Scan
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Benign
high confidence
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.
Assessment
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).

Like a lobster shell, security has layers — review code before you run it.

latestvk9703t33bfmztbmdvs3e7week183w77y
140downloads
0stars
2versions
Updated 4w ago
v1.0.1
MIT-0

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 <skill-name>-review-<date>.md

Comments

Loading comments...