LYGO SkillSpector

这是OpenClaw/ClawHub技能包的本地风险扫描工具，核心功能包括检测子进程调用、网络请求、硬编码密钥、恶意代码模式及声明匹配差异，支持单技能扫描、CI门禁（超阈值非零退出）、批量扫描、生成Markdown/HTML报告。适用于技能审计、安全安装检查、CI管控，输入为技能路径或包名，输出结果及报告，纯标准库实现，无网络/子进程调用

deepseekoracle

热度 16.09

最近更新 2026-08-13 09:44:01

分类：安全依赖：代码执行 / 读写文件应用场景：代码审查 / 报告撰写 / 自动化脚本

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add lygo-skill-spector

Metadata

name: lygo-skill-spector
description: LYGO SkillSpector — enhanced local pre-install skill risk scanner for OpenClaw/ClawHub packages. Scan, gate (CI max-band), batch, and Markdown report: subprocess/shell, network/HTTP/httpx, secrets, eval/exec, curl|bash, miners, claim mismatches. Use when auditing ClawHub skills, SkillSpector-style gates, safe install checks, malware triage. Pure stdlib. No network, no subprocess, no auto-install. Builder HTML/multi-root pack on SkillHub FULL if running a full stack. Install clawhub:@deepseekoracle/lygo-skill-spector.
version: 1.0.1
license: MIT-0
metadata: {"openclaw":{"emoji":"🔭","homepage":"https://github.com/DeepSeekOracle/lygo-protocol-stack/tree/main/docs/skills/lygo-skill-spector","requires":{"anyBins":["python","python3"]}},"lygo":true,"security":true,"audit":true,"clawhub_safe_install":true,"skillspector":true,"signature":"Delta9Phi963-SKILL-SPECTOR-v1.0.1","publisher":"deepseekoracle","clawhub":"https://clawhub.ai/deepseekoracle/skills/lygo-skill-spector","skillhub_full":"https://chatagent.ca/lygoskillhub.html#full-lygo","permissions":{"network":false,"shell":false,"subprocess":false,"filesystem":{"read":"skill directory path you pass to scan/gate/batch/report","write":"skill state/ only with --i-consent"},"publish":false,"auto_install":false}}

LYGO SkillSpector v1.0.1

Scan before you install — gate before you ship.

Enhanced local risk scanner for OpenClaw/ClawHub skill packages. Successor surface to Skill Gate with CI gate, batch, Markdown report, and deeper rule pack (httpx, curl|bash, PowerShell IEX, miners, HF/OpenAI key shapes).

Signature: Delta9Phi963-SKILL-SPECTOR-v1.0.1
ClawHub: @deepseekoracle/lygo-skill-spector
Audit response: references/SKILLSPECTOR_AUDIT.md (v1.0.0 meta-scan false positives fixed)

Builder version (full stack): If you run a full LYGO stack, the unlocked builder pack (HTML multi-root reports, multi-gate CI matrix, dashboard JSON) is on
SkillHub FULL LYGO
as lygo-skill-spector FULL RAW (+ builder/).

Dual channel

Channel	What you get
ClawHub (public)	`scan` · `gate` · `batch` · `report` · `self-demo` — tentacle
SkillHub FULL	Same + builder/ HTML batch, multi-root gate, CI summary — unlocked RAW

Install (public)

npx clawhub@latest install deepseekoracle/lygo-skill-spector

FULL engineer RAW: https://chatagent.ca/lygoskillhub.html#full-lygo

Commands

cd path/to/lygo-skill-spector
python scripts/self_check.py

# Scan one skill
python scripts/skill_spector.py scan "I:/E Drive/.grok/skills/lygo-context-guard"
python scripts/skill_spector.py scan lygo-ops-detector

# CI gate (exit non-zero if risk worse than max-band)
python scripts/skill_spector.py gate ./some-skill --max-band low

# Batch under a skills root
python scripts/skill_spector.py batch "I:/E Drive/.grok/skills"

# Markdown report
python scripts/skill_spector.py report ./some-skill
python scripts/skill_spector.py report ./some-skill --write last.md --i-consent

# Self-demo
python scripts/skill_spector.py self-demo

FULL builder only (SkillHub)

python builder/skill_spector_builder.py html-batch ./skills --write batch.html --i-consent
python builder/skill_spector_builder.py multi-gate ./skills --max-band elevated
python builder/skill_spector_builder.py ci-summary ./skills --write ci.json --i-consent

Exit	Meaning
0	clear / low (or under max-band)
5	elevated or claim mismatch
10	high / critical
2	bad path / not a skill

What it checks

subprocess / os.system / Popen / shell=True / PowerShell IEX
HTTP clients (urllib, requests, httpx, aiohttp), sockets, webhooks
curl|bash / wget|bash remote-code patterns
eval / exec, pickle, unsafe yaml
Hardcoded key-like / token-shaped strings (vendor project-key and HF-style prefixes)
Destructive deletes, force-push, auto-publish / ClawHub publish hints
Mining / keylogger-style IOC detection signals (rules only — not mining code)
Claim mismatch: frontmatter says network: false but code uses HTTP

Does not: download skills, install skills, execute scanned code, or phone home.

Pair with

Skill	Role
`lygo-skill-gate`	Lighter single-scan gate (still supported)
`lygo-context-guard`	Token budget + secret redact
`lygo-continuum`	Seal “scanned clean” as checkable claims
`lygo-kickstart-wizard`	Onboarding map

License

MIT-0.
Δ9Φ963 — verify before trust · local spector · human remains the publisher.