swarm
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from multiple subagents to generate a consolidated report, creating an attack surface for indirect prompt injection if worker outputs are influenced by malicious content.
- Ingestion points: Phase C in SKILL.md describes reading terminal results from background subagents.
- Boundary markers: The instructions lack specific delimiters or warnings to prevent the aggregator from obeying instructions embedded in worker reports.
- Capability inventory: Subagents are launched with general-purpose capabilities, including file system access via git worktrees and host agent tools.
- Sanitization: No sanitization or verification logic is specified for the reports before they are consolidated into the final output.
Audit Metadata