hypothesis-generation
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFENO_CODE
Full Analysis
- Data Exposure & Exfiltration (SAFE): The skill uses WebFetch and WebSearch tools to access non-whitelisted domains such as PubMed. This behavior is consistent with the primary purpose of scientific literature review and involves no access to sensitive local files or credentials.
- Indirect Prompt Injection (SAFE): The skill ingests data from external websites to inform hypothesis generation. Mandatory Evidence: 1. Ingestion: WebFetch and WebSearch (SKILL.md). 2. Boundary markers: Absent. 3. Capability inventory: Structured text generation via markdown templates. 4. Sanitization: Absent. The risk is minimal as the behavior is integral to the skill's research function.
- No Executable Code (SAFE): All files in the skill (SKILL.md, templates, and references) are markdown documents. No Python, Node.js, shell scripts, or binary executables are present.
Audit Metadata