Building Paper Screening Rubrics
Pass
Audited by Gen Agent Trust Hub on Oct 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes abstracts from PubMed, which are external data sources that could contain instructions designed to influence the agent's behavior. \n
- Ingestion points: Abstracts are fetched from eutils.ncbi.nlm.nih.gov and stored in test-set.json and abstracts-cache.json (SKILL.md). \n
- Boundary markers: There are no explicit delimiters or boundary markers defined to separate the untrusted abstract content from the agent's internal logic. \n
- Capability inventory: The skill is capable of executing shell commands (curl) and writing to the local file system (SKILL.md). \n
- Sanitization: No sanitization or validation of the retrieved abstract text is described before the scoring logic is applied. \n- [COMMAND_EXECUTION]: Executes curl commands to interact with the NCBI Entrez API for legitimate search and data retrieval tasks. \n- [EXTERNAL_DOWNLOADS]: Fetches scientific publication metadata and abstracts from the well-known and trusted National Center for Biotechnology Information (NCBI) service.
Audit Metadata