Jinguzhou is an early-stage safety gateway. Treat it as a defense layer, not a complete safety guarantee.

Please open a private security advisory or contact the maintainers before publishing details about vulnerabilities that could weaken policy enforcement, approval tokens, audit integrity, or tool-call controls.

• Keep approval secrets private.
• Use short approval token TTLs.
• Do not log full prompts in production unless you have a privacy review.
• Run destructive tools behind human review or explicit policy allowlists.