Bastion — AI Agent Safety GuardA local safety checkpoint for developer tools and AI coding agents. Designed, developed and maintained by Whistler Digital. Bastion reviews a proposed shell, Git, database or automation operation before it runs and returns one deterministic decision: Why Bastion?AI coding agents and automations can move quickly, but a single overly broad command can delete files, expose credentials or rewrite shared history. Bastion gives developers a readable second opinion before risky work reaches the operating system. It can identify operations involving:
Commands in VS CodeOpen the Command Palette with Bastion: Evaluate Risky OperationPaste or select a command and receive a local risk decision with the rules and reasons that matched. Bastion: Initialize ProjectCreate safe project defaults in Bastion: Run DiagnosticsInspect the current workspace integration and open a readable diagnostic report in the Bastion output channel. ExampleEvaluate a potentially destructive Git command:
Bastion returns More than a VS Code extensionThe same open-source policy engine is available as:
Quick CLI evaluation:
Privacy by design
Security boundaryBastion is a preflight policy layer, not an operating-system sandbox, antivirus product or complete prompt-injection defense. The calling tool must invoke Bastion and enforce its result. Use it together with least-privilege permissions, backups, code review and isolated execution environments. Open source and support
Released under the MIT License. Copyright Whistler Digital. |