Bartholomew Guard: shipping an AI safety gate for destructive actions

Description and use

Bartholomew Guard is a safety layer that blocks dangerous AI‑generated commands in development and CI environments, offering explanations and audit logs.

Problem

AI agents can issue destructive or high‑risk commands that bypass normal safety checks, leading to accidental data loss or system compromise.

Outcome

Implemented a practical safety layer that blocks unsafe AI‑generated shell commands, provides clear explanations, and logs decisions for auditability, enhancing developer confidence in AI‑assisted workflows.

Workflow

• 1. Install Bartholomew Guard as a VS Code extension or GitHub Action.
• 2. When an AI agent proposes a shell command, the guard evaluates intent and pattern matching.
• 3. If the command is flagged as destructive, the guard blocks execution and returns a denial message.
• 4. The decision is logged in a receipt‑style record for audit.
• 5. Developers continue the workflow with safe commands.
• 6. Review logs to understand denied actions and adjust policies if needed.

Evidence

We built Bartholomew to stop unsafe AI-driven commands before they execute.

Source

The guard evaluates command intent, flags destructive patterns, and denies actions like shell deletion, risky file operations, and unsafe execution paths.

Source