drug-drug
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data retrieved during the literature search phase, which could potentially contain malicious instructions or biased content designed to influence the agent's clinical report.
- Ingestion points: Untrusted data enters the context via
web_searchandweb_fetchoperations targeting PubMed, CrossRef, and general web sources (SKILL.md). - Boundary markers: The skill does not explicitly define delimiters to separate retrieved external text from its internal instructions.
- Capability inventory: The agent is restricted to information retrieval and report generation; it lacks access to sensitive local files, system commands, or administrative privileges.
- Sanitization: The skill mitigates risks by requiring multi-source cross-validation and applying a formal evidence-grading framework to verify retrieved information.
- [EXTERNAL_DOWNLOADS]: Fetches clinical information, publication metadata, and FDA labels from authoritative research platforms including PubMed (NCBI), CrossRef, and DailyMed (NLM).
Audit Metadata