We built Bartholomew to stop unsafe AI-driven commands before they execute. The core idea is simple: if an agent tries to run something destructive, high-risk, or out of policy, the action is blocked before it reaches the shell.
This started as a real enforcement problem, not a demo. The guard evaluates command intent, flags destructive patterns, and denies actions like shell deletion, risky file operations, and unsafe execution paths. It also records a receipt-style outcome so the decision is auditable.
We packaged it into a VS Code extension and a GitHub Action so it works in both developer workflows and automation pipelines. That gives teams a practical way to enforce guardrails without waiting for a broad platform rollout.
The result is a concrete layer of safety for AI-assisted development: block the dangerous action, explain why it was denied, and keep the rest of the workflow moving safely.
We’re still in the early days of agentic tooling, but the important step is to make the safety layer real, visible, and enforceable.
Top comments (0)