Skill Vetter

Installing AI agent skills without vetting can introduce security vulnerabilities such as data breaches or system compromises. Skill Vetter is a security-first vetting protocol designed to systematically assess skill risks before deployment.

Problem Context

AI skill platforms (e.g., ClawdHub or GitHub) offer a wide range of shared skills, but sources vary in quality. Directly installing unknown skills may lead to credential theft, permission abuse, or malicious code execution, increasing operational and security burdens. Traditional installation flows often skip proactive checks, leaving risks unaddressed.

Core Capabilities and Workflow

Skill Vetter implements comprehensive vetting through structured steps:
- Source Check: Verify publisher reputation and repository status, such as prioritizing official skills or high-star repos (1000+ Stars).
- Code Review (Mandatory): Read all skill files, checking for red flags like accessing credentials, modifying system configs, or involving trading operations.
- Permission Scope: Analyze requested permissions to ensure minimal privilege and avoid over-authorization.
- Risk Classification: Determine actions based on risk levels (🟢 LOW, 🟡 MEDIUM, 🔴 HIGH, ⛔ EXTREME), with high-risk skills requiring human approval.

After vetting, a standardized report is generated with risk levels and recommendations. The protocol also defines a trust hierarchy to guide scrutiny intensity from official skills to unknown sources.

Applicability and Considerations

Skill Vetter is applicable for pre-installation security assessments but is not a universal defense:
- When risks are uncertain, seek human decisions to avoid auto-installing high-risk skills (e.g., those involving security configs or root access).
- Vetting records aid future reference but may need adjustment for specific environments.
- The protocol emphasizes: no skill is worth compromising security—when in doubt, do not install.