swarm
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from multiple subagents which constitutes an attack surface.
- Ingestion points: Aggregates terminal results from N workers in Phase C.
- Boundary markers: Missing explicit delimiters to separate worker output from the main agent report.
- Capability inventory: Can spawn subagents and read local configuration files.
- Sanitization: No specific sanitization mentioned for worker reports before consolidation.
- [COMMAND_EXECUTION]: The skill manages the lifecycle of multiple subagent processes.
- Evidence: Phase B instructs the agent to spawn N subagents with run_in_background: true and subagent_type: generalPurpose.
Audit Metadata