skill-safety-verifier_skill

Overview

This skill performs security-first vetting for AI skills before installation. It analyzes code, dependencies, and permission requests to surface red flags and produce a clear risk classification. The output includes a risk radar, warnings, and actionable recommendations to guide install decisions.

How this skill works

The verifier clones or extracts the skill and runs parallel scans: network/socket checks, static code pattern detection, and vulnerability lookups (e.g., GitHub Advisory API or local cache). It aggregates socket alerts, vulnerability severity, and permission scope into a numeric risk score and classifies risk level with tailored guidance. Results are presented as a concise report with remediation steps and an install recommendation.

When to use it

• Before installing any third-party skill from ClawdHub, GitHub, or other repositories
• When an existing installed skill requests expanded permissions or network access
• During CI/CD pipelines to block high-risk skill changes before deployment
• When auditing archived skill versions or restoring backups
• Prior to granting skills persistent or system-level privileges

Best practices

• Always run the verifier before installation — automated checks catch common issues quickly
• Treat the report as guidance: review flagged code manually, especially for command execution or eval usage
• Enforce least privilege: only grant the minimal scopes required for functionality
• Run high-risk skills inside isolated sandboxes or containers if you must proceed
• Use caching and non-blocking advisory lookups to keep install workflows responsive

Example use cases

• Auto-scan a new skill fork during a CI job and fail the build on high-risk findings
• Pre-install check integrated into an agent marketplace to present safety ratings to users
• Audit archived SkillHub repositories to identify legacy vulnerabilities before reactivation
• Verify updates to a skill that introduces network calls or command execution
• Run targeted scans on installed skills after detecting suspicious outbound traffic