freeze
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from external sources ('outside-the-repo state') including third-party services, databases, and live logs.
- Ingestion points: Instructions in
SKILL.mddirect the agent to 'read each named fact live' from external sources and bring them into the active conversation context. - Boundary markers: The instructions mandate that 'observed facts and the open gaps' be 'kept apart,' providing a structural boundary, but the skill lacks specific delimiters or instructions to ignore potential commands embedded within the external data.
- Capability inventory: The skill invokes other agent capabilities (
clarify,draft, andjudgeskills) to process, format, and validate the collected information. - Sanitization: The skill relies on a 'judge' skill to verify that facts are not 'invented,' but it does not specify any sanitization, filtering, or escaping techniques to prevent instructions within the external data from influencing the agent's behavior.
Audit Metadata