swarm
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill coordinates sub-workers whose output is aggregated, which creates a potential surface for indirect prompt injection if workers ingest untrusted data.
- Ingestion points: Worker results and diff evidence aggregated in SKILL.md Phase C.
- Boundary markers: Includes explicit instructions to 'Inspect each result's diff or source evidence' and 'Do not paste raw worker dumps'.
- Capability inventory: The skill provides coordination logic and does not utilize high-privilege shell or network tools directly.
- Sanitization: Instructions require the agent to 'run the relevant check before accepting' and to provide a consolidated summary instead of raw data.
Audit Metadata