AI Skill Security Vetter

Problem

Before installing unknown skills from ClawdHub, GitHub repositories, or other agents, teams often lack a consistent security review process. A skill may request credentials, hidden network access, system commands, or configuration changes. Skill Vetter treats “vet before install” as a hard constraint and helps decide whether a skill should be loaded and what level of human approval it requires.

How it works

• Source check: evaluate trust first; official OpenClaw skills may use lower scrutiny but still require review, high-star repos and known authors get moderate review, and new or unverified sources receive maximum scrutiny.
• Code review: read all files in the skill and look for red flags such as data exfiltration, hidden network requests, credential access, security config changes, shell execution, or root access.
• Permission scope: assess the required file, browser, API, system, or credential access to avoid over-authorization.
• Risk classification: classify the skill as LOW, MEDIUM, HIGH, or EXTREME; LOW needs basic review, MEDIUM requires full code review, HIGH requires human approval, and EXTREME should not be installed.
• Vetting report: produce a structured conclusion that can be reused for audits and team discussion.

Caveat: this skill is best used as a pre-install security gate for unknown code from ClawdHub, GitHub, or other agents. The provided material does not include the full red-flag checklist or quick vet command details, so real use should still inspect the complete skill files, repository history, dependencies, and runtime environment. Skills requesting credentials, trading, system permissions, or root access should always be escalated to a human.