wizard
Pass
Audited by Gen Agent Trust Hub on Aug 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructions allow for the ingestion of external data which could contain malicious instructions designed to influence agent behavior.
- Ingestion points: Step 6 in
SKILL.mddirects the agent to verify facts from external "authoritative documentation" or the human user. - Boundary markers: The skill does not define delimiters or boundary markers to separate ingested external content from its internal operational logic.
- Capability inventory: The skill includes the capability to generate persistent operational artifacts ("maintained operations path") in the repository as described in Step 7 of
SKILL.md. - Sanitization: No explicit methods for sanitizing, escaping, or validating the content retrieved from external sources are prescribed.
Audit Metadata