medical-research-gap-finder
Pass
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill inherently possesses an attack surface for indirect prompt injection because it is designed to ingest and analyze external, untrusted content from medical literature and preprint servers.
- Ingestion points: Research data and metadata are retrieved from external sources during the literature review phase (Step 2 in SKILL.md).
- Boundary markers: The skill implements robust boundary markers through its 'Hard Rules' and 'Pseudo-Gap Rejection Rules,' which mandate specific output structures and verification of metadata.
- Capability inventory: The skill lacks dangerous capabilities such as arbitrary code execution, filesystem modification, or credential access that could be exploited via injection.
- Sanitization: The skill employs a verification protocol (defined in references/literature-retrieval-and-citation.md) that requires cross-referencing persistent identifiers (DOIs, PMIDs) to ensure data integrity.
- [EXTERNAL_DOWNLOADS]: The skill references and retrieves data from well-known scientific databases including PubMed, Google Scholar, Web of Science, and arXiv. These are established services and their use is necessary for the skill's primary research purpose.
- [SAFE]: Static analysis of the skill's logic, metadata, and reference materials revealed no malicious patterns. The instructions prioritize scientific integrity, explicitly prohibit clinical decision-making, and do not include any obfuscated code or unauthorized execution patterns.
Audit Metadata