multi-agent-coordinator

Pass

Audited by Gen Agent Trust Hub on Mar 21, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill implements an orchestration layer that processes and stores data from multiple sub-agents, creating a surface for indirect prompt injection.\n
  • Ingestion points: Untrusted data enters the orchestrator's context via run.py through the --record-output, --handoff, and --register commands, which process arbitrary strings for tasks, roles, and results.\n
  • Boundary markers: The state management logic lacks explicit delimiters or instructions that would prevent the agent from misinterpreting stored sub-agent data as system-level instructions.\n
  • Capability inventory: The run.py script is restricted to local file system operations (reading and writing state.yaml) and does not directly execute subprocesses or perform network operations, which limits the immediate impact of an injection.\n
  • Sanitization: No input validation or sanitization is applied to the strings received from sub-agents before they are stored in the shared state file.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 21, 2026, 05:55 AM