folklore-variant-evidence
Pass
Audited by Gen Agent Trust Hub on Sep 11, 2026
Risk Level: SAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill makes network requests to an external, non-whitelisted domain (api.helena.bio) to retrieve variant evidence and publication details. While the instructions strictly prohibit sending sensitive patient context, phenotypes, or identifiers, the connection to this third-party service represents a potential exfiltration vector if the agent is misused.
- [INDIRECT_PROMPT_INJECTION]: The skill processes scientific results and literature corpus data from external databases, creating an indirect prompt injection surface.
- Ingestion points: The agent ingests data through tools such as search_variant_evidence, search_literature_corpus, and get_publication_details.
- Boundary markers: The skill enforces clear input boundaries to refuse sensitive patient data, but does not provide explicit markers or instructions for the agent to disregard potential instructions embedded within the retrieved scientific evidence.
- Capability inventory: The agent uses retrieved data (status codes) to drive workflow branching and chains canonical keys and publication identifiers into downstream tool calls.
- Sanitization: No explicit content sanitization or validation logic is defined for the data returned from the Folklore MCP API.
Audit Metadata