Agent Gate: Open-source tool blocks AI agents from taking dangerous actions

Developer WildeConforce released agent-gate, a Python library that inserts a deterministic policy gate between an AI model's output and real-world actions. The gate issues single-use tokens only after explicit code-level checks, preventing prompt injection or reasoning errors from causing irreversible harm. It includes features like pre-action rejection, sandbox backups, and tamper-evident audit logs.