recall
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill acts as an aggregator for large amounts of historical and external data, creating a surface for indirect instructions to influence agent behavior.
- Ingestion points: Data is ingested through
t3_thread_read(chat history) and thewhyskill (source control, issue trackers, and error logs). - Boundary markers: The instructions explicitly mandate that subagents be given 'read-only' status with 'no writes' and 'no git commands' allowed. It also includes an instruction to 'sanitize private context before any public output'.
- Capability inventory: The agent has the ability to spawn sub-tasks via
delegate_taskand executegitandghcommands. - Sanitization: No specific automated sanitization or schema enforcement is defined for the content of the threads or issues being analyzed.
Audit Metadata