doublecheck
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes arbitrary, untrusted target text or prior AI outputs to perform factual auditing and claim extraction. If the analyzed text contains embedded instructions designed to deceive the model, it could influence the verification logic or prompt interpolation during web searches.
- Ingestion points: Target text passed by the user or from previous agent outputs is parsed dynamically in
SKILL.mdduring the Layer 1 self-audit phase. - Boundary markers: Absent; the skill lacks explicit encapsulation instructions or guidelines directing the agent to ignore command sequences inside the text being evaluated.
- Capability inventory: Incorporates the
web_searchtool to dynamically look up queries generated from the claims. - Sanitization: Lacks explicit filtering or escaping mechanisms for the text properties prior to extracting queries.
Audit Metadata