informed-patient
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted health data provided by users and external information retrieved via web searches to generate a medical evidence review.
- Ingestion points: User-provided symptoms, medical history, and specific health claims in Phase 1 (SKILL.md, references/phase-1-interview.md); external medical literature retrieved during the structured search in Phase 2 (SKILL.md, references/phase-2-search.md).
- Boundary markers: The agent follows a rigid three-phase sequence and uses a specific output template (references/output-template.md) to isolate and structure the content.
- Capability inventory: The skill utilizes web search capabilities to fetch literature and file-system write operations to deliver the artifact (SKILL.md).
- Sanitization: No explicit sanitization or escaping of external content is defined before it is interpolated into the generated report or subsequent prompts.
- [COMMAND_EXECUTION]: The skill instructs the agent to write a structured markdown report to the local filesystem.
- Evidence: The SKILL.md file specifies the creation of a file named health-evidence-review-[condition]-[YYYY-MM-DD].md in the working directory.
- [EXTERNAL_DOWNLOADS]: The skill retrieves information from external medical and government databases via web search queries.
- Evidence: The search strategy in references/literature-search-strategy.md identifies Cochrane, PubMed, NICE, and AHRQ as primary sources for evidence retrieval.
- Context: These sources are well-known, official health organizations and research databases.
Audit Metadata