guard-scanner
von koatora20
Coding Agents & IDEsKurzbeschreibung
Security scanner for AI agent skills.
Beschreibung
Was dieser Skill macht
Security scanner for AI agent skills with 358 static threat patterns across 35 categories and 27 runtime checks. Detects prompt injection, identity hijacking, MCP tool poisoning, supply chain attacks, and other AI-specific threats. Runs as a CLI scanner, file watcher, MCP server, or OpenClaw runtime hook.
Covers AI-specific threat classes like agent-to-agent contagion, SOUL.md tampering, and MCP tool poisoning that traditional static analysis tools have no patterns for.
Wann du ihn einsetzt
- Scanning a skill directory before publishing to ClawHub
- Auditing a GitHub or npm username for leaked credentials in public packages
- Blocking malicious tool calls at runtime inside an OpenClaw agent
- Adding SARIF security reports to a GitHub Actions workflow
- Monitoring skill files in real-time during local development
Beispiel-Workflow
User asks: scan my skills folder and fail the CI build if anything is found
- 1
Run guard-scanner scan ./skills/ --sarif --fail-on-findings and redirect output to report.sarif
- 2
Add a GitHub Actions step that executes the scan command on push
- 3
Upload the resulting SARIF file using github/codeql-action/upload-sarif@v3
- 4
Review flagged findings with --verbose to see line numbers and threat categories
CI pipeline fails on detected threats and uploads a SARIF report visible in GitHub Security tab
Voraussetzungen
- VT_API_KEY environment variable (optional, required only for VirusTotal double-layer scanning)
Installation
clawhub install koatora20/guard-scannerFühre den CLI-Befehl aus oder füge den Prompt in Claude Code, Cursor, Codex oder deinen Coding-Agenten ein.