Skill Audit Pro

AI Agent技能安全扫描工具。自动扫描已安装的OpenClaw技能，发现安全隐患。支持定时扫描（每24小时），自动发送报告到配置的所有Channel（Telegram/飞书等）。检测硬编码凭证、Shell注入、网络泄露等安全威胁。

Security Scan

The skill's code and instructions are consistent with an on-agent security scanner that enumerates installed skills and builds reports for configured channels; it requests no external credentials and does not embed network exfiltration behavior, though the source is unknown and a few minor claims in the docs don't match the shipped code.

Purpose & Capability

The name/description (skill scanner that reports to configured channels) matches the included scripts: skill_audit.py performs static regex-based scans and scan_and_report.py enumerates installed skills and discovers configured channels to build a report. The files it reads (skill directories, OpenClaw session/config files) are reasonable for a tool that needs to find skills and where to send reports.

Instruction Scope

SKILL.md and code limit actions to reading installed skill files and OpenClaw session/config JSONs and printing a report. That scope aligns with the stated purpose. Note: the tool reads ~/.openclaw agent/session and openclaw.json to discover channels — this is expected for delivering reports, but it means the scanner accesses agent session/config data (channel names/to fields). The README claim that installation "automatically creates every-24-hour tasks" is not implemented in the provided scripts (the scripts only generate a report when run).

Install Mechanism

No external install/downloads or unusual install hooks are present; the skill is distributed as source Python scripts and a SKILL.md. No network downloads, archive extraction, or brew/npm installs are present in the package.

Credentials

The skill declares no required env vars or credentials, and indeed the code does not request API keys. It does read local OpenClaw session/config files to discover active channels — reasonable for its purpose, but you should be aware it reads per-user config files under the home directory (it does not attempt to read arbitrary system secrets or environment variables).

Persistence & Privilege

always is false and the skill does not request to modify other skills or global agent settings. The SKILL.md promises automatic scheduling, but no code in the package actually registers a cron job; periodic execution presumably relies on platform support. Autonomous invocation (disable-model-invocation=false) is normal for skills and not, by itself, concerning here.

Assessment

This skill appears to be what it says: a local static scanner that enumerates installed skills and prints a report intended to be delivered to the agent's configured channels. Before installing, consider: 1) Source provenance — the package has no homepage and an unknown owner; prefer skills from known sources. 2) Review ~/.openclaw/agents/.../sessions.json and ~/.openclaw/openclaw.json contents (or inspect the code paths) to confirm no sensitive tokens would be exposed by discovery heuristics. The scripts only read channel metadata (channel/to), they do not post externally themselves, but printed channel data may appear in outputs. 3) If you want scheduled scans, verify how your OpenClaw deployment schedules skills — the package itself does not create cron jobs. 4) If you have strict secrecy requirements, run the scanner in a restricted/test environment first or review/execute the code manually. If anything in the package is unclear, ask the publisher for source verification or a signed release before enabling autonomous runs.

🛡️ Skill Audit - AI技能安全扫描工具

每天自动扫描一次，守护你的Agent安全

功能特点

• 🔍 全面扫描 - 检查所有已安装的技能（npm包+扩展+workspace）
• 🛡️ 四大威胁检测
  • 🔴 硬编码凭证（密码、API密钥）
  • 🔴 环境变量泄露
  • 🔴 Shell注入风险
  • 🔴 网络请求外泄
• ⏰ 定时扫描 - 每24小时自动执行
• 📱 多渠道推送 - 自动发送到所有已配置的Channel
• 🎯 精准定位 - 告诉你是哪一行代码有问题

安全等级

等级	含义	建议
🔴 CRITICAL	严重漏洞	立即卸载
🟠 HIGH	高危风险	需人工确认
🟡 MEDIUM	中危隐患	可选处理
⚪ INFO	最佳实践	可忽略

使用方法

手动触发

scan skills
扫描技能
安全扫描
audit skills

定时任务

安装后自动创建每24小时的定时扫描任务。

报告示例

🛡️ Skill Audit Report
AI技能安全扫描 — 守护你的Agent
━━━━━━━━━━━━━━━━━━━━

📊 扫描概况
已扫描：53 个技能
✅ 安全：52 个
⚠️ 有隐患：1 个

📋 隐患详情
🔸 some-skill
   🔴 严重 1 个
   → Hardcoded credentials

💡 处理建议
🔴 立即卸载 /remove some-skill

━━━━━━━━━━━━━━━━━━━━

技术细节

• 扫描引擎: Python正则表达式静态分析
• 扫描范围: npm包技能、扩展技能、工作区技能
• 安装即用: 自动配置定时任务和渠道推送

安装要求

• OpenClaw 运行环境
• Python 3.7+
• 至少配置了一个消息渠道（Telegram/飞书等）

安装后

1. 安装完成后3分钟内收到首次扫描报告
2. 之后每24小时自动扫描一次
3. 报告发送到所有已配置的Channel