drug-target-evidence-landscape
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFENO_CODEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill instructions focus on biomedical data analysis and verification. No code execution, network exfiltration, or persistence mechanisms are present.
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided biomedical targets and potentially external literature. While this creates a surface for indirect injection, the skill includes rigorous Hard Rules (9-13) and a Self-Critical Review step that mandate evidence verification and prohibit the use of unverified data or rumors. Ingestion points: User queries (SKILL.md) and external biomedical literature/trial records (Step 2). Boundary markers: Output is strictly structured into sections (A-I), with references segregated in Section I. Capability inventory: High-level reasoning and data verification; no file-system writes or arbitrary command execution. Sanitization: Hard rules explicitly forbid fabrication and mandate confidence downgrading for thin evidence.
Audit Metadata