conduct-research
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from web search results and primary documents, creating an attack surface where malicious content could attempt to influence the agent's behavior.
- Ingestion points: External URLs and primary documents retrieved during the iterative search and discovery phases (Phases 1-3).
- Boundary markers: Implements a formal "Citation Directed Acyclic Graph" (Citation DAG) structure to prevent circular reasoning and a "Chain-of-Verification" (CoVe) protocol to isolate and verify individual claims.
- Capability inventory: The skill is permitted to use
write_file,create_file, andexecute_command(for tools likecurlandc2patool). - Sanitization: Employs a strict source hierarchy (Tier 1-4) that strictly forbids citing AI-generated content and mandates empirical triangulation across at least three independent primary sources.
Audit Metadata