SKILL.md — skill-vetter-plus

Name

skill-vetter-plus

Description

Security scanner for AI agent skills. Performs static analysis on skill manifests, bundled scripts, and prompt content to surface common vulnerabilities and anti-patterns before installation or publication.

Version

1.0.0

Author

CertainLogic

License

MIT

Category

security, scanner, audit

Tags

security, vulnerability, scanner, static-analysis, skill-audit, prompt-injection

Requirements

• Python 3.10+
• Optional: semgrep (for extended rule coverage)

Installation

Place the skill-vetter-plus/ folder into your OpenClaw skills directory and restart.

Usage

CLI

# Basic scan
openclaw skill vetter-plus scan <path-to-skill>

# Verbose output with remediation hints
openclaw skill vetter-plus scan --verbose <path-to-skill>

# JSON output for CI integration
openclaw skill vetter-plus scan --json <path-to-skill>

# Scan multiple skills
openclaw skill vetter-plus scan --batch skill1/ skill2/ skill3/

# Use extended semgrep rules (if installed)
openclaw skill vetter-plus scan --deep <path-to-skill>

As a Library

from vetter import SkillScanner

scanner = SkillScanner()
report = scanner.scan("/path/to/skill")
for finding in report.findings:
    print(f"[{finding.severity}] {finding.rule}: {finding.message}")

Configuration

Create vetter-config.yaml in the skill directory to customize rules:

rules:
  secret_detection: true
  prompt_injection: true
  unsafe_shell: true
  network_exfiltration: true
  file_permissions: true

severity_threshold: medium   # ignore low-severity findings
max_file_size_mb: 5
exclude_patterns:
  - "*.min.js"
  - "node_modules/**"

Output Format

Findings include:

• severity: critical / high / medium / low / info
• rule_id: identifier for the check that triggered
• file: relative path to the problematic file
• line: line number (if applicable)
• message: human-readable description
• remediation: suggested fix (when available)

Limitations

• Static analysis only; cannot detect runtime behavior
• Rule coverage is best-effort, not exhaustive
• May produce false positives on benign patterns
• Extended deep scanning requires separate semgrep installation

Changelog

1.0.0

• Initial release
• Core scanner engine with 25 built-in rules
• CLI with verbose and JSON output modes
• Batch scanning support