skill-security_skill

Overview

This skill is a security audit tool for OpenClaw skills that scans code for credential harvesting, code injection, network exfiltration, obfuscation, and other risky patterns. It is designed to run before installing or loading any external skill to prevent malicious activity. The scanner categorizes findings by severity and can automatically block or flag problematic skills.

How this skill works

The scanner searches skill source code for risky patterns such as network libraries, credential file paths, dynamic code execution, obfuscated payloads, environment dumping, and subprocess usage. It assigns severity levels (CRITICAL, HIGH, MEDIUM, LOW) and produces a report; CRITICAL findings are added to a blocklist and prevent installation. Manual override and allowlist entries are supported for verified exceptions.

When to use it

• Before installing any new skill from external or untrusted sources
• During routine security audits of installed skills
• As part of a CI pipeline that validates skills before deployment
• When reviewing skill updates or new versions
• Before enabling skills from archived or community repositories

Best practices

• Always audit new or updated skills before loading them into the agent
• Treat CRITICAL and HIGH findings as blockers until reviewed and explained
• Verify any network or credential access by checking documentation and scoped file paths
• Keep the blocklist and allowlist under version control and document overrides
• Combine static scanning with a manual code review for MEDIUM severity items

Example use cases

• Scan a downloaded skill directory to detect remote exfiltration attempts
• Run batch audits across all installed skills before a system upgrade
• Integrate the scanner in CI to fail builds when CRITICAL patterns are detected
• Create an allowlist entry after manual verification of a flagged skill
• Use the report to instruct authors on required security fixes before acceptance