skill-security-framing
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a framework for managing untrusted data ingestion. 1. Ingestion points: External URLs and content fetched via tools like WebFetch. 2. Boundary markers: Explicitly defined 'SECURITY CONTEXT' and 'UNTRUSTED CONTENT' delimiters are required for all analysis workflows. 3. Capability inventory: Identifies subagent analysis and external fetching as primary risk surfaces. 4. Sanitization: Enforces protocol validation, RFC 1918 blocklists for SSRF prevention, and content truncation limits to mitigate payload risks.
- [EXTERNAL_DOWNLOADS]: References fetching external data via WebFetch and uses a transform for Twitter URLs to utilize a specialized API (fxtwitter.com), accompanied by strict validation logic to ensure only public HTTPS resources are accessed.
- [SAFE]: The skill is a security-focused documentation and guideline package that provides mitigations against common AI agent attack vectors.
Audit Metadata