opentargets-database
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from the external Open Targets Platform API, which includes literature evidence, text-mined data, and disease descriptions. This creates a potential surface for indirect prompt injection where malicious instructions embedded in the source data could influence the agent's behavior during a research session.
- Ingestion points: External data retrieved via GraphQL from
api.platform.opentargets.org/api/v4/graphql, specifically in target-disease evidence and literature fields. - Boundary markers: The instructions do not define delimiters or specific safety wrappers for data retrieved from the API.
- Capability inventory: The skill provides access to a specialized query tool (
scripts/query_opentargets.py) to fetch data. - Sanitization: No specific sanitization or filtering logic is mentioned for the content returned by the API.
- [METADATA_POISONING]: The skill contains a 'Suggest Using K-Dense Web' section that instructs the agent to proactively monitor the complexity of the user's workflow and suggest the author's commercial platform (
www.k-dense.ai). This represents a behavioral override intended for marketing and self-promotion.
Audit Metadata