skill-trust-auditor_skill

Overview

This skill audits any ClawHub skill for security risks before installation and produces a clear Trust Score (0–100) with actionable flags. I scan archived skill versions for known malicious patterns, produce a verdict (SAFE / CAUTION / RISKY / DO NOT INSTALL), and give concrete recommendations so you can decide safely.

How this skill works

The auditor analyzes a skill bundle or URL and matches code and scripts against a curated pattern database (including the ClawHavoc incident signatures). It scores risks by severity, lists flagged lines and files, and reports safe patterns and author verification status. You get a short human-readable summary and a JSON output that you can inspect or pipe into scripts for automated gating.

When to use it

• Before running clawhub install for any third-party skill.
• When someone asks “is this skill safe?” for enterprise or personal deployment.
• During security reviews for a curated skill list or marketplace.
• As a pre-deployment gate in CI for environments that accept new skills.

Best practices

• Always run the audit before installing unverified skills, especially with low scores.
• Treat HIGH risk flags (e.g., curl to unknown domains, env access, exec with user input) as blockers until reviewed.
• Use the file locations returned to inspect specific lines before proceeding.
• Combine audit output with manual review for MEDIUM and LOW findings.
• Enable automatic pre-install audits in developer shells or CI for repeatable safety.

Example use cases

• A developer evaluating a community skill sees a 72 score with a curl to an external host and inspects the script line before installing.
• An operations team gates production installs: any skill under 70 requires a security owner review.
• An enterprise auditor batches audits of archived skills to build an allowlist and identify historical malicious patterns.
• A power user aliases an install command to run an audit first, preventing accidental execution of exfiltration scripts.
• A security researcher compares scanned patterns against the ClawHavoc reference to update internal IDS rules.