swarm
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill provides a structured workflow for agent orchestration without any detected malicious patterns, unauthorized network access, or credential exposure.
- [INDIRECT_PROMPT_INJECTION]: The skill incorporates an aggregation workflow that processes data from external worker agents.
- Ingestion points: Terminal results from N parallel workers are read in SKILL.md (Phase C: Aggregate).
- Boundary markers: The instructions explicitly state 'Do not paste raw worker dumps' and require a 'compact result table,' which provides a structural delimiter for processed data.
- Capability inventory: The skill uses create_agent, create_workspace, list_profiles, and list_models for orchestration.
- Sanitization: Results are consolidated by the parent agent into one-line evidenced issues rather than raw data ingestion, relying on the model's summarization to filter potential injection content.
Audit Metadata