swarm
Pass
Audited by Gen Agent Trust Hub on Nov 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow for aggregating data from multiple parallel worker agents, which creates a potential surface for indirect prompt injection. The skill includes mitigations such as instructing the agent to use structured status reports and avoid including raw worker output in the final report.\n
- Ingestion points: Data enters the context when results are read from worker agents via agent:// protocols.\n
- Boundary markers: The skill uses a 'done predicate' and specific result statuses (PASS, ISSUES, BLOCKED) to delimit content.\n
- Capability inventory: The skill manages workers via the task orchestration tool, writes evidence to local files (local://), and can terminate processes via proc:// protocols.\n
- Sanitization: Explicit instructions are provided to avoid pasting raw worker dumps into the final consolidated report.\n- [DATA_EXFILTRATION]: The instructions reference platform configuration paths (~/.omp/agent/config.yml) to explain model roles and overrides. This access is part of the agent platform's internal orchestration and does not involve sending sensitive data to external domains.
Audit Metadata