scientific-problem-selection
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- Indirect Prompt Injection Surface: The skill includes an ingestion surface for untrusted data. 1. Ingestion points: User-provided research ideas (SKILL.md) and literature search results from external sources like PubMed (01-intuition-pumps.md). 2. Boundary markers: The instructions do not define explicit delimiters or 'ignore' instructions for these inputs. 3. Capability inventory: The skill uses standard conversational capabilities and implied tool access for searches, but does not include file-writing or code execution scripts. 4. Sanitization: There are no specified filtering or sanitization steps for external data. This surface is managed by the agent's core guardrails and the collaborative nature of the skill.
- Academic and Service Alignment: The skill recommends using well-known services like PubMed for literature validation and incorporates established scientific frameworks, aligning with standard research practices.
Audit Metadata