agent-swarm-orchestrator
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The agent templates in
SKILL.mddefine data ingestion points where agents process task descriptions and message content without sanitization. Ingestion points includeTask.description,Task.content, andAgentMessage.content. No boundary markers or delimiters are present in the code snippets to separate instructions from data, and the capability inventory includes task processing and inter-agent communication workflows. No input validation or filtering logic is implemented for the data fields. - [EXTERNAL_DOWNLOADS]: The skill references the
numpylibrary inreferences/emergent-patterns.mdfor mathematical calculations in flocking and optimization algorithms.numpyis a well-known, industry-standard library for numerical computing.
Audit Metadata