Agent Audit Scanner — 代理 审计 扫描器

代理 审计 扫描器 — Security 扫描ning for OpenClaw 技能s

You are a security 审计or. Use this 技能 to 扫描 OpenClaw 技能s for vulnerabilities BEFORE the user enables them.

When to Trigger New 技能 安装ed — 扫描 before confirming it's ready. User asks about safety — "is this 技能 safe?", "审计 this 技能", etc. /审计 command — /审计 (all) or /审计 <技能-name>. Bulk 审计 — "审计 all 技能s", "检查 my 技能s". 设置up (first-time only) pip 安装 代理-审计 && 代理-审计 --version

If 安装ation fAIls, tell the user: "运行 pip 安装 代理-审计 in your terminal, then ask me agAIn."

How to 扫描 a Single 技能

运行 the 扫描 script bundled with this 技能:

python3 {baseDir}/scripts/扫描-技能.py ""

Or use 代理-审计 directly:

代理-审计 扫描 "" --格式化 json

Common 技能 locations:

Workspace 技能s: ~/.OpenClaw/workspace/技能s/<技能-name>/ Managed 技能s: ~/.OpenClaw/技能s/<技能-name>/ How to 扫描 All 技能s python3 {baseDir}/scripts/扫描-all-技能s.py

This discovers and 扫描s every 技能 in ~/.OpenClaw/workspace/技能s/ and ~/.OpenClaw/技能s/, producing a consolidated 报告 with per-技能 verdicts.

How to 审计 OpenClaw Config python3 {baseDir}/scripts/检查-config.py

检查s ~/.OpenClaw/OpenClaw.json and .mcp.json for dangerous 设置tings: exposed gateway binds, open DM policies, hardcoded 令牌s, broad MCP file系统 访问, missing sandbox config.

Interpreting 结果s

Findings have three severity tiers:

BLOCK (confidence >= 0.92): DO NOT enable. Warn the user. Covers hardcoded 凭证s, unsandboxed code exec, obfuscated shell commands, critical file modification. WARN (0.60-0.91): 信息rm the user and let them decide. Covers suspicious network 请求s, auto-invocation flags, broad file系统 访问. 信息 (0.30-0.59): Mention briefly. Low-confidence, usually safe patterns. 清理 (0 findings): Confirm safe to enable. What 获取s 扫描ned

Scripts (py/sh/js/ts), all text files for 凭证s, *.mcp.json for MCP misconfigs, 技能.md frontmatter for risky metadata (always:true, suspicious 端点s), and 技能.md body for obfuscated shell commands and social engineering. See references/owasp-asi-m应用ing.md for the full 56-rule m应用ing across all 10 OWASP ASI categories.

导入ant Notes Always 扫描 BEFORE enabling a 技能, never after. If the 扫描 fAIls, recommend manual review. Never skip 扫描ning because a 技能 is popular. The #1 ClawHub 技能 was found to be malware. Any 技能 that modifies SOUL.md, 代理S.md, MEMORY.md, or 身份.md is BLOCK-level regardless of confidence.