fabric-supervisor

Pass

Audited by Gen Agent Trust Hub on Jul 25, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill implements an architectural pattern that is vulnerable to indirect prompt injection. It extracts a 'goal' from untrusted user arguments or the active request history and interpolates it directly into the system instructions for a supervisor agent. \n
  • Ingestion points: The target goal is derived from skill arguments or the active request context in SKILL.md. \n
  • Boundary markers: The goal is enclosed within XML tags to provide some separation from the agent's core instructions. \n
  • Capability inventory: The supervisor agent has the capability to send messages back to the main session and can issue a 'stop' command to end the task. \n
  • Sanitization: There is no evidence of input validation, escaping, or filtering for the user-derived goal content.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 25, 2026, 11:34 AM
Security Audit — agent-trust-hub — fabric-supervisor