research-recipes
Pass
Audited by Gen Agent Trust Hub on Apr 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: Executes shell commands via bash to perform network requests and data processing using curl, python3, and R.
- [EXTERNAL_DOWNLOADS]: Retrieves research data from established scientific APIs and databases including the National Center for Biotechnology Information (NCBI), OpenAlex, Semantic Scholar, and cBioPortal.
- [PROMPT_INJECTION]: Configures the agent to automatically execute multi-step research pipelines without requesting human confirmation upon detecting specific keyword triggers.
- [PROMPT_INJECTION]: Vulnerable to indirect prompt injection due to the processing of untrusted content (paper abstracts and full-text) from external search results without explicit sanitization or boundary markers.
- [DATA_EXFILTRATION]: Performs network operations to external scientific services to transmit search queries and retrieve genomic data for analysis.
Audit Metadata