multi-agent-supervisor

Pass

Audited by Gen Agent Trust Hub on May 15, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill integrates a 'System Agent' with capabilities for Bash and shell execution. This is a powerful feature intended for orchestration but represents an execution surface that could be abused if the supervisor's logic is compromised.
  • [DATA_EXFILTRATION]: The architecture includes an 'Email Agent' capable of SMTP/IMAP operations. While legitimate for its purpose, this provides a potential channel for data exfiltration if sensitive information is routed through the agent in an uncontrolled manner.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8). It ingests 'agent_outputs' which may contain untrusted data from external sources (e.g., via the Browser Agent). These outputs are then used in 'result_synthesis' and can influence subsequent 'task_decomposition' and 'agent_routing' decisions. The documentation does not specify the use of strict boundary markers or sanitization logic to prevent embedded instructions in these outputs from being obeyed by the supervisor.
Audit Metadata
Risk Level
SAFE
Analyzed
May 15, 2026, 05:16 PM
Security Audit — agent-trust-hub — multi-agent-supervisor