librarian
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill researches and downloads external content, which poses a risk of ingesting malicious instructions from third-party sources. * Ingestion points: The skill gathers research from the web and saves local copies in a source-files directory (SKILL.md). * Boundary markers: It uses a structured template defined in SOURCE-FORMAT.md to compartmentalize research data, including specific headings for intent, scope, and evidence. * Capability inventory: The subagent performs network requests to gather data and write operations to save Markdown files and source content to the local filesystem. * Sanitization: The skill contains explicit instructions to reject executables and archives, verify MIME types, and ensure that downloaded content is never executed (SKILL.md).
Audit Metadata