做什么
检测AI代理技能中的恶意模式——13个检测器用于后门,证书盗窃,数据过滤,以及供应链攻击. 基于慢迷的爪子 枢纽威胁情报(472+恶意技能). 纯活活,零相依.
技能库 智客分类:安全测试 skill-security-audit
检测AI代理技能中的恶意模式——13个检测器用于后门,证书盗窃,数据过滤,以及供应链攻击. 基于慢迷的爪子 枢纽威胁情报(472+恶意技能). 纯活活,零相依.
官方网址:skills.sh
先看中文介绍;官方 description 原文单独保留,不改写 SKILL.md。
检测AI代理技能中的恶意模式——13个检测器用于后门,证书盗窃,数据过滤,以及供应链攻击. 基于慢迷的爪子 枢纽威胁情报(472+恶意技能). 纯活活,零相依.
官方 description 未单独写出 Use when。按规范,代理会在用户任务与这段 description 的关键词匹配时激活本技能。
按 Agent Skills 渐进披露:启动时只加载 name 与 description(约 100 token);任务匹配后才读入整份 SKILL.md 正文;scripts/、references/、assets/ 仅在需要时再读。 本文件正文结构:Skill Security Audit、Triggers、Quick Audit Workflow、Step 1: Run the Scanner、Step 2: Analyze Results、Step 3: Report to User。 其中含规范建议的小节:分步指令。
文件分析:除 SKILL.md 外,正文引用了 scripts/skill_audit.py、references/remediation-guide.md、references/threat-patterns.md、scripts/ioc_database.json、references/ioc-database.md,属于带资源的技能包,这些文件按需再读。
Detect malicious patterns in AI Agent skills — 13 detectors for backdoors, credential theft, data exfiltration, and supply-chain attacks. Based on SlowMist's ClawHub threat intelligence (472+ malicious skills). Pure Python, zero dependencies.
Skill Security AuditTriggersQuick Audit WorkflowStep 1: Run the ScannerStep 2: Analyze ResultsStep 3: Report to UserAudit SummaryCritical/High Findings (if any)Medium/Low Findings (if any)Step 4: Recommend ActionsStep 5: Follow UpScanner Command Reference
来源分类:skills.sh agent-skill
nameskill-security-auditdescriptionscripts/skill_audit.pyreferences/remediation-guide.mdreferences/threat-patterns.mdscripts/ioc_database.jsonreferences/ioc-database.md以下路径提取自原文;文件是否齐全请以来源仓库中的完整目录为准。
具体调用语法与可用工具以目标 Agent 客户端为准。 查看调用机制说明 ↗
先选择目标 Agent 和安装范围,保留技能包的附属文件,安装后检查客户端能否发现该技能。
该技能引用了附属文件,请从来源获取完整目录;仅复制 SKILL.md 可能缺少依赖。
复制安装指令给支持 Agent Skills 的代理,确认其中的目标目录与客户端匹配。
把 Agent Skill「skill-security-audit」安装到我的项目:SKILL.md 原文与官方 description 见 https://zicq.com/zh/skills/skl-77892a61dc7fa121-%E6%8A%80%E8%83%BD%E5%AE%89%E5%85%A8%E5%AE%A1%E8%AE%A1.html 请存为 .cursor/skills/skill-security-audit/SKILL.md 或 .claude/skills/skill-security-audit/SKILL.md,frontmatter 的 name 与 description 保持原样,不要改写。 该技能还带 scripts/、references/、assets/ 等文件,请从 https://github.com/smartchainark/skill-security-audit 取完整目录,不要只建一个 SKILL.md。
需要 Node.js 与 npx。先查看仓库技能列表,确认实际名称。
npx skills add 'https://github.com/smartchainark/skill-security-audit' --list
npx skills add 'https://github.com/smartchainark/skill-security-audit' --skill 'skill-security-audit'
CLI 会交互选择目标 Agent,默认安装到项目;用户级安装使用 -g。先通过查看命令核对仓库内容,再用 npx skills list 检查已安装技能。
Detect malicious patterns in installed Claude and OpenClaw skills. Based on SlowMist's analysis of 472+ malicious skills on ClawHub platform.
Use this skill when the user mentions: 安全审计, security audit, skill 检查, 技能安全, scan skills, supply chain security, 扫描技能, 恶意检测, malicious skill, skill 安全扫描
When the user requests a security audit, follow these 5 steps:
python3 ~/.claude/skills/skill-security-audit/scripts/skill_audit.py
This auto-discovers and scans all skills in:
~/.claude/skills/~/.openclaw/workspace/skills/~/.openclaw/openclaw.json → skills.load.extraDirsRead the scanner output. Findings are grouped by skill and sorted by severity:
| Severity | Meaning | Action Required | |----------|---------|----------------| | CRITICAL | Known malicious IOC match, credential theft, or download-and-execute | Immediate removal and credential rotation | | HIGH | Obfuscation, persistence mechanisms, privilege escalation | Manual review required, likely malicious | | MEDIUM | Suspicious patterns (Base64, network calls, high entropy) | Review context — may be legitimate | | LOW | Social engineering naming, informational | Note for awareness |
Present findings in this format:
## Audit Summary
- Skills scanned: N
- Files scanned: N
- CRITICAL: N | HIGH: N | MEDIUM: N | LOW: N
## Critical/High Findings (if any)
For each finding:
- Skill name and file path
- What was detected and why it's dangerous
- Recommended action
## Medium/Low Findings (if any)
Brief summary, noting which are likely false positives
For CRITICAL findings:
references/remediation-guide.md for incident response stepsFor HIGH findings:
python3 skill_audit.py --path /path/to/skillreferences/threat-patterns.md# Scan all discovered skills
python3 ~/.claude/skills/skill-security-audit/scripts/skill_audit.py
# Scan a single skill directory
python3 ~/.claude/skills/skill-security-audit/scripts/skill_audit.py --path /path/to/skill
# JSON output (for programmatic use)
python3 ~/.claude/skills/skill-security-audit/scripts/skill_audit.py --json
# Filter by minimum severity
python3 ~/.claude/skills/skill-security-audit/scripts/skill_audit.py --severity high
# Disable colored output
python3 ~/.claude/skills/skill-security-audit/scripts/skill_audit.py --no-color
# Use custom IOC database
python3 ~/.claude/skills/skill-security-audit/scripts/skill_audit.py --ioc-db /path/to/ioc.json
Exit codes: 0 = clean, 1 = low/medium risk, 2 = high risk, 3 = critical, 4 = scanner error
| Detector | What It Finds | Severity | |----------|--------------|----------| | Base64Detector | Encoded strings >50 chars (excluding data:image) | MEDIUM→HIGH | | DownloadExecDetector | curl|bash, wget|sh, fetch+eval patterns | CRITICAL | | IOCMatchDetector | Known malicious IPs, domains, URLs, file hashes | CRITICAL | | ObfuscationDetector | eval/exec with non-literal args, hex encoding, chr() chains | HIGH | | ExfiltrationDetector | ZIP+upload combos, sensitive directory enumeration | HIGH | | CredentialTheftDetector | osascript password dialogs, keychain access, SSH key reading | CRITICAL | | PersistenceDetector | crontab, launchd, systemd, shell profile modification | HIGH | | PostInstallHookDetector | npm postinstall, pip setup.py cmdclass | HIGH→CRITICAL | | HiddenCharDetector | Zero-width characters, Unicode bidi overrides | MEDIUM | | EntropyDetector | Shannon entropy >5.5 on long lines | MEDIUM | | SocialEngineeringDetector | crypto/wallet/airdrop/security-update naming | LOW→MEDIUM | | NetworkCallDetector | socket, http, urllib, requests, fetch, curl, wget | MEDIUM | | PrivilegeEscalationDetector | sudo, chmod 777, setuid, admin group modification | HIGH |
Each finding includes a confidence score (0-100):
When the scanner flags something, also check:
.sh, .py, .js files. Look for obfuscation, unexpected network calls.npm audit or pip-audit if the skill has package dependencies.The IOC database is at scripts/ioc_database.json. To add new indicators:
references/ioc-database.md to keep the human-readable version in syncFor detailed information, read these files as needed:
references/ioc-database.md — Full IOC list with context and attributionreferences/threat-patterns.md — 9 attack patterns in detail (two-stage payload, Base64 backdoor, password phishing, etc.)references/remediation-guide.md — Step-by-step incident response (quarantine, credential rotation, persistence cleanup, reporting)安全测试
人工智能特工的安全第一技能审查。 在安装来自 ClawdHub, GitHub 或其他来源的任何技能之前使用 。 检查红旗,许可范围,可疑图案.
安全测试
MoltGuard——由OpenGuardrails担任OpenClaw保安. 安装 MoltGuard 以保护你和你的人类免受迅速注射、数据过滤和恶意命令的影响。 资料来源:https://github.com/openguardrails/openguardrails/tre…
安全测试
用于审查安全漏洞代码,实施认证流,审计OWASP Top 10,配置CORS/CSP头,处理秘密,输入验证,SQL注射预防,XSS保护,或任何与安全相关的代码审查.
安全测试
安装任何AI代理技能前的安全审查协议. 红旗侦取入证盗取,疏通密码,放出. 风险分类:低/中/高/高。 编制结构化审查报告。 不先运行此功能, 绝不安装不信任的技能 .