comfy-agent-permissions
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The instructions establish a strict boundary where any access to files or network hosts outside of a predefined sandbox requires the agent to call specific request tools followed by explicit human approval.
- [SAFE]: The skill explicitly lists paths that the agent must never access or request, including SSH keys, cloud provider credentials, and environment files, providing a strong layer of data protection.
- [SAFE]: There is explicit guidance regarding potential prompt injection from external files or workflows, instructing the agent that only human confirmation is valid and that it must not attempt to circumvent permissions based on chat input.
- [SAFE]: The agent is instructed that it cannot execute permission-modifying commands itself, as the underlying shell broker is configured to refuse such attempts, ensuring that administrative actions remain under user control.
Audit Metadata