skills/aipoch/medical-research-skills/single-compound-network-toxicology-disease-link-reference-grounded/Gen Agent Trust Hub
single-compound-network-toxicology-disease-link-reference-grounded
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill instructions and associated reference files define a robust analytical workflow without any executable malicious code or dangerous instructions.
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided compound and disease identifiers, creating an attack surface for indirect prompt injection. This is mitigated by: (1) Ingestion points: Input fields for chemical compounds and disease phenotypes in SKILL.md; (2) Boundary markers: Mandatory section headers (A-J) and standardized output templates; (3) Capability inventory: The skill acts as a text-based planner and does not invoke external tools or file operations; (4) Sanitization: Mandatory Dataset Disclaimers and Self-Critical Risk Reviews that categorize outputs as hypothesis-generating rather than factual instructions.
- [METADATA_POISONING]: The skill package contains a JSON evaluation report that includes self-referential safety claims. While these are documented, they do not affect the skill's operational security or logic.
Audit Metadata