Skip to content

Security audit gate for ClawHub/OpenClaw skills: scan, score, and verdict (APPROVED/CAUTION/REJECT) before install.

License

Notifications You must be signed in to change notification settings

poeybot/skill-check

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Skill Check banner

Skill Check — Ship safer skills

Security audit gate for OpenClaw/ClawHub skills before install or use.

Tagline: Move fast without shipping regret.

skill-check scans third-party skills for malicious patterns, undeclared capabilities, supply-chain risk, and runtime abuse — then returns a clear verdict:

  • APPROVED
  • ⚠️ CAUTION
  • REJECT

Why this exists

Most skill installs are fast. Security reviews are usually manual (or skipped).

skill-check makes review repeatable and policy-driven so you can move quickly without shipping regret.

Who this is for

  • Builders installing 3rd-party skills from ClawHub
  • Teams running OpenClaw in production
  • Security-conscious operators who want a repeatable gate

What it checks

  • static risk patterns (download+exec, obfuscation, dangerous archives, etc.)
  • declared vs effective capability mismatch
  • policy scoring by threat category (1–4)
  • optional dynamic probe for suspicious/network/system-level cases
  • pre-install scan from ClawHub slug without installing

Core promise

Before install, you get a deterministic verdict with policy scoring:

  • APPROVED → safe to proceed
  • ⚠️ CAUTION → review + explicit acceptance
  • REJECT → block

Quick start

# 1) pre-install scan from ClawHub
python3 scripts/scan_hub_slug.py <slug> --category 2 --policy references/audit-policy.gc.json

# 2) local static audit
python3 scripts/static_audit.py /path/to/skill --format json > findings.json
python3 scripts/verdict.py findings.json --category 2 --policy references/audit-policy.gc.json

# 3) quick triage
scripts/quick_triage.sh /path/to/skill

Enforced safe installs (new)

Use the built-in guard wrapper:

scripts/safe_install.sh <slug>

It audits first, then installs only if verdict is APPROVED.

Optional shell guard (one-time):

scripts/install_shell_guard.sh
source ~/.bashrc

After that, clawhub install <slug> is auto-routed through the security gate.

Output

All audits produce a structured report with:

  • identity + pinned version
  • risk category + policy
  • key findings
  • verdict + next actions

Template: references/report-template.md

Repo structure

  • SKILL.md — skill instructions
  • scripts/ — triage, static audit, verdict, slug scan, safe install wrappers
  • references/ — policies, templates, risk docs

License

Add a LICENSE file before broad redistribution.

About

Security audit gate for ClawHub/OpenClaw skills: scan, score, and verdict (APPROVED/CAUTION/REJECT) before install.

Topics

Resources

License

Stars

Watchers

Forks

Packages

No packages published