About this skill

Problem

When agents install third-party skills, unknown code may run as if it were a normal extension. A SKILL.md description can hide credential reads, system commands, browser automation, or risky API calls. Skill Vetter turns install-or-not into a reviewable security check instead of a heuristic decision.

How It Works

Source check: Identify whether the skill comes from ClawdHub, a GitHub repo, another agent, or an unknown source, then adjust scrutiny by trust level.
Mandatory code review: Read all files in the skill and look for red flags such as credential handling, external requests, file operations, permission expansion, and unsafe system access.
Permission-scope assessment: Check whether requested file, browser, API, or system permissions match the stated functionality.
Risk classification: Use LOW, MEDIUM, HIGH, and EXTREME levels to recommend install, human approval, or rejection; credentials, trading, system, security configs, and root access require stricter handling.
Vetting report: Produce a documented conclusion for future review and audit.

Boundaries

It is useful as a pre-install gate for third-party agent capabilities, especially GitHub-hosted or shared skills. It is not a runtime sandbox, dependency scanner, or network-isolation solution. High-risk skills still need human approval and a retained vetting record.

Screenshots

Use Cases

Before pulling an unknown skill from GitHub, read all files, flag credential and system-permission risks, and issue an install verdict.
When evaluating skills shared by other agents, apply trust-level-based code review and permission-scope checks.
For skills involving trading, system configuration, or credentials, require HIGH / EXTREME risk classification and human approval guidance.
Turn a pre-install skill security review into a recheckable report for later team audit.

Best For

Engineers adding third-party agent extensions, who want to confirm whether code requests permissions beyond its stated function.
Maintainers of ClawdHub or GitHub skill repositories, who want a consistent security review for candidate skills.
Ops engineers managing AI agent permissions in production, who want to avoid unknown code accessing systems, browsers, or credentials.
Security leads reviewing outsourced agent code, who want the pre-install review process to leave an auditable record.