multi-agent-patterns
Pass
Audited by Gen Agent Trust Hub on Apr 15, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The instructions and code provide educational material and utilities for building multi-agent systems. No malicious behavior such as credential theft, data exfiltration, or unauthorized command execution was detected.
- [SAFE]: External references to LangGraph, AutoGen, CrewAI, and ArXiv point to trusted repositories and official documentation.
- [INDIRECT_PROMPT_INJECTION]: The skill describes architectures that process untrusted data (user queries) through multiple agents, which constitutes an indirect prompt injection surface.
- Ingestion points: User queries provided to supervisor or swarm nodes as documented in SKILL.md and references/frameworks.md.
- Boundary markers: Not explicitly defined in code snippets, though 'explicit coordination protocols' are recommended in the instructional text.
- Capability inventory: Includes file system read/write access via the FileSystemCoordination class in references/frameworks.md and general tool use such as web searching referenced in SKILL.md.
- Sanitization: Recommended through 'validation' steps and the use of 'verification agents' described in the failure mode mitigations section of SKILL.md.
Audit Metadata