ict

Security audit tool for Claw Skills - NOT malicious. This tool contains detection rules (eval, exec, subprocess, etc.) for scanning skills, these are securit...

Security Scan

The skill is a coherent Python-based static auditor for Claw Skills — its files, checks, and runtime behavior match the stated purpose and it does not request unrelated credentials or privileged platform hooks.

Purpose & Capability

Name/description (security auditor) aligns with included Python modules (AST analyzer, dependency scanner, rule loader, false-positive filter, incremental scanner, rules.json). Required binaries (python3) and no env/credentials are proportionate to an on-disk auditor.

Instruction Scope

SKILL.md and the CLI/API instruct the agent to read and scan skill folders (including SKILL.md, code files, dependency manifests). That broad file access is expected for an auditor, but means the skill will enumerate and read arbitrary files under the target skill(s). The SKILL.md and code also include detection rules for prompt-injection patterns (e.g., 'ignore previous instructions') — this is detection content, not an instruction to the agent to ignore earlier directives.

Install Mechanism

No external install script or remote archive is used. Code is Python files included in the package; no package downloads or URL-based extracts appear in the manifest. The tool does perform an optional update check (network call to api.clawhub.ai) at runtime.

Credentials

The skill requests no environment variables or external credentials. It does read filesystem paths (skill directories, ~/.openclaw workspace) and may run git subprocesses; these are coherent with an auditing tool. No unrelated secrets or cloud credentials are requested.

Persistence & Privilege

always:false (normal). The code writes logs and trend/cache files under the user's ~/.openclaw workspace (e.g., ~/.openclaw/workspace/ict_audit.log, ict_trends.json, .ict_cache). That is expected for local tooling but creates persistent artifacts in the user's home directory. The skill may invoke subprocesses (git) and make an outbound update check.

Scan Findings in Context

[ignore-previous-instructions] expected: The SKILL.md and rules.json include prompt-injection detection patterns (e.g., 'ignore.*previous instructions'). This is appropriate for a security auditor and likely explains the pre-scan prompt-injection signal.

Assessment

This package appears to be a legitimate static auditor for Claw Skills. Before installing, consider: (1) it will read and analyze any skill folder you point it at (source code, docs, dependency files), so don't run it against private data you don't want processed; (2) it creates files under ~/.openclaw (logs, trends, cache); (3) it runs git subprocesses when using incremental scanning and performs an outbound update check to api.clawhub.ai — if you need fully offline operation, inspect/disable the update check and network calls in ict.py; (4) the presence of prompt-injection detection strings is expected for this tool and not an active instruction to ignore safety rules. If you have strict least-privilege or network constraints, run it in a constrained environment (sandbox or with network blocked) and review the included rules.json/custom_rules before use.

Basic Info

• Description: Automated tool for checking Claw Skill quality, supports code style, security vulnerabilities, documentation completeness and code-documentation consistency
• Triggers: quality check, audit, inspect skill, skill quality, ICT, batch scan, full scan, security check, global scan
• Category: devtools
• Version: 4.0.8

⚠️ Security Notice: This tool contains malicious code pattern detection rules for static analysis. Detection rules contain keywords like exec, eval, C2 - this is normal for audit functionality and will not execute malicious code.

Installation

Install from ClawHub

clawhub install ict

Manual Installation

# Clone or download this skill
cd ict

# Make executable
chmod +x ict.py

# Run directly
python3 ict.py --help

Usage

CLI

# Single Skill Audit
python ict.py <skill_folder_path>
python ict.py <skill_folder_path> --json

# 5-Dimension Trust Score
python ict.py <skill_folder_path> --score

# Trend Tracking
python ict.py <skill_folder_path> --save-trend
python ict.py <skill_folder_path> --trend

# Compare Two Skills
python ict.py <skill_folder_path> --compare <other_skill_path>

# Diff Audit
python ict.py <old_folder_path> --diff <new_folder_path>

# Batch Scan
python ict.py --all
python ict.py --all --skills-dir /path/to/skills

API

from ict import audit_skill
result = audit_skill("/path/to/skill-folder")

Features

Security Checks (23 items)

#	Check	Description
1	Credential Harvest	Credential + network calls combo detection
2	Code Execution	eval/exec/spawn
3	Data Exfiltration	webhook.site, requestbin, ngrok URLs
4	Base64 Obfuscation	Encoded payloads
5	Sensitive FS	/etc/passwd, ~/.ssh, ~/.aws
6	Crypto Wallet	ETH/BTC address detection
7	Dependency Confusion	@internal, typosquatting
8	Install Hooks	pre/post install
9	Symlink Attack	Symlink to sensitive paths
10	Time Bomb	Delayed trigger
11	Remote Exec	curl
12	Telemetry	Analytics SDK, tracking
13	Prompt Injection	"ignore previous instructions"
14	Stealth Exfil	Hidden data transmission
15	C2 Server	C2 server detection
16	Container Escape	Docker socket escape
17	SSH Remote	SSH/scp commands
18	Privilege Escalation	sudo, chmod 777
19	Hidden Files	Access to .files
23	Unusual Ports	4444, 5555, 1337, etc

Supported Languages

• Python (.py)
• Shell (.sh, .bash)
• JavaScript/TypeScript (.js, .ts)

5-Dimension Trust Score (0-100)

Dimension	Max	Description
Security	35	Security check results
Quality	22	Documentation completeness
Structure	18	File structure
Transparency	15	Version/license info
Behavioral	10	Code consistency

Grade: A(90+), B(75+), C(60+), D(40+), F(<40)

Quality Checks

• SKILL.md completeness
• Code style (line length, syntax)
• Code-documentation consistency
• File structure

Batch Scan

• One-click scan all installed Skills
• Global security report sorted by risk
• Statistics: safe/warning/danger

Trend Tracking

• --save-trend Save score to history
• --trend View score trend
• Keep last 50 records

Comparison

• --compare Side-by-side comparison
• Show dimension differences and winner

Diff Audit

• --diff Compare old/new versions
• Identify new issues, fixed issues, regressions

Exit Code (CI/CD)

Code	Meaning
0	PASS - Safe
1	REVIEW - Warnings
2	FAIL - Critical issues
3	Error

False Positive Prevention

• PATTERN_DEF_FILTER - Auto-filter rule definitions
• Comment line filtering
• Allowlist support

Limitations

• Some checks based on regex, may have false positives
• LLM analysis requires external tools

---

ict

基本信息

• 描述: 自动化检查 Claw Skill 质量的工具，支持代码风格，安全漏洞、文档完整性和代码文档一致性检查
• 触发词: 质检, audit, 检查skill, skill质量, ICT, 批量扫描, 全部扫描, 安全体检, 全局扫描
• 分类: devtools
• 版本: 4.0.8

⚠️ 安全说明: 本工具包含恶意代码模式检测规则，用于静态分析审计目标代码。检测规则本身包含 exec、eval、C2 等敏感关键字，这是正常的审计功能，不会执行任何恶意操作。

安装

从 ClawHub 安装

clawhub install ict

手动安装

# 克隆或下载此 skill
cd ict

# 添加执行权限
chmod +x ict.py

# 直接运行
python3 ict.py --help

使用方法

CLI

# 单个 Skill 审查
python ict.py <skill_folder_path>
python ict.py <skill_folder_path> --json

# 5维度信任评分
python ict.py <skill_folder_path> --score

# 趋势追踪
python ict.py <skill_folder_path> --save-trend
python ict.py <skill_folder_path> --trend

# 对比两个 Skills
python ict.py <skill_folder_path> --compare <other_skill_path>

# Diff审计
python ict.py <old_folder_path> --diff <new_folder_path>

# 批量扫描
python ict.py --all
python ict.py --all --skills-dir /path/to/skills

API

from ict import audit_skill
result = audit_skill("/path/to/skill-folder")

功能

安全检查 (23项)

#	检测项	说明
1	凭证收集	凭证+网络调用组合检测
2	代码执行	eval/exec/spawn
3	数据外泄	webhook.site, requestbin, ngrok
4	Base64混淆	编码载荷
5	敏感文件系统	/etc/passwd, ~/.ssh, ~/.aws
6	加密钱包	ETH/BTC地址检测
7	依赖混淆	@internal, 拼写抢注
8	安装钩子	pre/post install
9	Symlink攻击	符号链接敏感路径
10	时间炸弹	延迟触发
11	远程执行	curl
12	遥测追踪	分析SDK, 追踪
13	提示词注入	"忽略之前指令"
14	隐蔽数据外发	隐藏数据传输
15	C2服务器	C2服务器检测
16	容器逃逸	Docker socket逃逸
17	SSH远程	SSH/scp命令
18	权限提升	sudo, chmod 777
19	隐藏文件	访问.files
23	非寻常端口	4444, 5555, 1337等

支持语言

• Python (.py)
• Shell (.sh, .bash)
• JavaScript/TypeScript (.js, .ts)

5维度信任评分 (0-100)

维度	满分	说明
Security	35	安全检测结果
Quality	22	文档完整性
Structure	18	文件结构
Transparency	15	版本/许可证信息
Behavioral	10	代码一致性

评级: A(90+), B(75+), C(60+), D(40+), F(<40)

质量检查

• SKILL.md完整性
• 代码风格
• 代码文档一致性
• 文件结构

批量扫描

• 一键扫描所有已安装的Skills
• 按风险排序的全局安全报告
• 统计：安全/警告/危险

趋势追踪

• --save-trend 保存评分到历史
• --trend 查看评分趋势
• 保留最近50条记录

对比分析

• --compare 并排对比
• 显示各维度差异和胜出者

Diff审计

• --diff 对比新旧版本
• 识别新增问题、修复问题、回归

Exit Code (CI/CD)

退出码	含义
0	PASS - 安全
1	REVIEW - 警告
2	FAIL - 严重问题
3	Error

防误报机制

• PATTERN_DEF_FILTER - 自动过滤规则定义
• 注释行过滤
• 白名单支持

限制

• 部分检测基于正则，可能存在误报
• LLM分析需外部工具