Show HN: Deterministic execution gate for AI agents (github.com)

A new GitHub library named “execution-gate” introduces a deterministic execution framework for AI agents, requiring explicit permission (ALLOW) before any actions with side effects are executed. This is a significant advance in agent frameworks, which typically operate with default execution, potentially risking unintended consequences. The library emphasizes security and transparency by defaulting to a "fail-closed" status—blocking any action unless a policy is explicitly defined to allow it. All decisions are logged in structured JSON format, enhancing observability and auditability for developers. The execution-gate library uses YAML-based policy rules for easy configuration and provides decorator-based enforcement to ensure decisions are made before any execution occurs. Key features include deterministic decision-making (no probabilistic outcomes), a clear method for checking intents, and built-in mechanisms for handling rule violations. With its single-purpose design, the library promotes minimal integration overhead and avoids framework lock-in, making it a valuable tool for developers looking to implement secure and predictable behavior in AI agents.