swarm
Pass
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates a multi-agent workflow where a parent agent aggregates results from multiple 'workers'. This structure creates a surface for indirect prompt injection; if a worker processes untrusted external data and includes it in its report, it could influence the parent agent's subsequent logic during the aggregation and reporting phases.
- Ingestion points: Phase C (Aggregate) involves reading terminal results and reports from N parallel workers.
- Boundary markers: Absent. The instructions do not prescribe the use of delimiters or explicit 'ignore embedded instructions' warnings for the content provided by workers.
- Capability inventory: The skill enables the creation of general-purpose subagents in cloud and local environments, filesystem operations (writing to /tmp/ or branches), and consolidated reporting.
- Sanitization: Absent. While the instructions suggest formatting (one-line summaries), there is no requirement for technical sanitization, validation, or escaping of the ingested worker data.
Audit Metadata