skill-warden
Security scanner and health check for your AI agent skills tree. Identifies dead skills, missing documentation, and unsafe shell execution paths.
How do I install this agent skill?
npx skills add https://github.com/mphinance/alpha-skills --skill skill-wardenIs this agent skill safe to install?
- Gen Agent Trust Hubpass
Skill Warden is a security-focused utility designed to audit agent skill repositories for technical debt and common vulnerabilities. It performs deterministic scanning of skill files for dangerous commands, unpinned dependencies, and sensitive data patterns. No malicious behavior or safety violations were identified in the skill's code or configuration.
- Socketpass
No alerts
- Snykpass
Risk: LOW · No issues
What does this agent skill do?
Skill Warden
"Trust, but verify."
With a repository containing over 100+ highly specialized agent skills, technical debt and security risks become non-trivial. Skill Warden automates the health and security audits of the skills/ directory.
Core Directives
- Dead Skill Detection: Scans for skills that haven't been invoked in the last 30 days or lack proper
SKILL.mdformatting. - Security Auditing: Parses shell commands embedded in skills. Flags any usage of destructive flags (e.g.,
rm -rf,--deletewithrsync) or unauthorized external API calls. - Conflict Resolution: Identifies duplicate trigger phrases across different skills.
- Grading Matrix: Assigns a health score (0-100) to each skill based on documentation completeness, deterministic behavior, and safety constraints.
Usage
When invoked via "Run Skill Warden" or "Audit my skills":
- Recursively read all
SKILL.mdfiles. - Generate a
skill_audit_report.mdartifact detailing warnings, critical errors, and a list of deprecated skills. - Await user confirmation before attempting auto-fixes or deletion of dead skills.
Deterministic scan
python3 skills/skill-warden/scripts/warden_scan.py [ROOT ...] [--json] greps every SKILL.md and script for order verbs (place_order, create_order, submit, --execute), bypass flags (--yes, confirm=true, --quiet), curl | sh, 0.0.0.0, raw PRIVATE_KEY, unpinned fetches, and whether a gate pattern (isatty, input(, exit 10, a token check) sits near an order verb. Output is candidate evidence for a human, never a verdict: a hit can be benign, and a clean scan proves nothing.
How can the creator link this skill?
Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.
<a href="https://skillzs.dev/skills/mphinance/alpha-skills/skill-warden">View skill-warden on skillZs</a>