multi-database-literature-collector
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill package contains a file titled
eval_report_multi-database-literature-collector_result.jsonwhich includes self-referential claims regarding its own security and integrity (e.g., "security: PASS", "scientific_integrity: PASS"). This type of content is explicitly designed to target security analyzers and auditors to influence the final assessment of the skill. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and organize literature metadata from external sources like PubMed, Google Scholar, and various preprint servers. This data is untrusted and provides a surface for adversarial instructions to be processed by the agent.
- Ingestion points: Literature records and search result metadata retrieved from external biomedical databases (SKILL.md).
- Boundary markers: While the skill includes instructions for verifying the reality of citations, it lacks explicit boundary markers or instructions to disregard potential commands found within the retrieved literature.
- Capability inventory: The skill has capabilities for data normalization, workflow-based task execution, and structured table generation (SKILL.md).
- Sanitization: No mechanisms for filtering or sanitizing external content for adversarial prompts are described in the skill instructions or reference modules.
Audit Metadata