tooluniverse-binder-discovery
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from external literature sources such as PubMed, BioRxiv, and MedRxiv. These platforms host third-party content which could theoretically contain instructions designed to influence the agent's behavior during the summary or analysis phase.
- Ingestion points: Data retrieved via
PubMed_search_articles,BioRxiv_search_preprints, andMedRxiv_search_preprintsis incorporated into the research report. - Boundary markers: The skill uses a structured report template but does not explicitly define delimiters or instructions to ignore embedded prompts within the retrieved literature.
- Capability inventory: The agent has the capability to write local files (Markdown reports, CSV compound lists) and call a large library of specialized research tools (e.g., NVIDIA NIM, ChEMBL, PubChem).
- Sanitization: The instructions do not specify a validation or sanitization step for the natural language content retrieved from external databases.
- [EXTERNAL_DOWNLOADS]: The workflow relies on data retrieved from several well-known and established scientific databases and technology services, including UniProt, ChEMBL, RCSB PDB, BindingDB, PubChem, and NVIDIA's NIM API. These interactions are standard for the skill's stated purpose of biological research.
Audit Metadata