swarm
Pass
Audited by Gen Agent Trust Hub on Nov 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill aggregates output from potentially untrusted subagents into a final report, creating an injection surface. \n
- Ingestion points: Worker terminal results are read during Phase C (Aggregation) and Phase D (Reporting) in SKILL.md. \n
- Boundary markers: The instructions lack specific boundary markers or delimiters for separating subagent content from system instructions. \n
- Capability inventory: The skill utilizes the subagent spawning tool and reads terminal output. \n
- Sanitization: The skill instructs the agent to consolidate results into a table and one-liners rather than dumping raw output, which provides a degree of summarization-based sanitization.
Audit Metadata