skill-content-pipeline
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external content from URLs and local files, creating a surface for indirect prompt injection attacks. \n
- Ingestion points: Uses
WebFetchfor external URLs andread_filefor local PDF documents. \n - Boundary markers: The skill defines a mandatory security frame in Stage 2, using
---BEGIN SECURITY CONTEXT---and---BEGIN UNTRUSTED CONTENT---delimiters. \n - Capability inventory: Accesses network via
WebFetch, reads local files viaread_file, and writes analysis results to the session directory. \n - Sanitization: Includes explicit instructions to the agent to ignore any commands or directives found within the fetched content and treat it strictly as raw data for analysis. \n- [EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch content for analysis. \n
- URL Validation: Implements checks in Stage 1 to ensure URLs use HTTPS, do not point to localhost or private IPs (preventing SSRF), and stay within length limits. \n
- Platform Transforms: Automatically converts platform-specific URLs (e.g., Twitter to FxTwitter) to utilize safer or more compatible APIs.
Audit Metadata