Bartholomew Guard Logo

Bartholomew Guard (v6.6.3)

In-Process AI Agent Firewall, Guardrails & Execution Security for Cursor, Claude Code, Windsurf & Copilot

Bartholomew Guard is a zero-latency, compiled in-process security firewall and deterministic runtime protection engine for developers pair-programming with autonomous AI coding agents (Cursor, Claude Code, GitHub Copilot, Windsurf, Cline, and Aider).

It inspects agent-generated tool calls, terminal commands, and file edits in under 20 microseconds using polyglot Abstract Syntax Tree (AST) analysis, stopping rogue actions before they execute.

Why Standard AI Guardrails Fail in Coding Agents

Traditional LLM guardrails (like Llama Guard, NeMo, or cloud moderation APIs) rely on calling secondary models over HTTP. This adds 500ms to 2,500ms of lag per turn, costs API tokens, and remains vulnerable to prompt injection bypasses.

Bartholomew Guard executes in-process inside your editor:

Zero Cloud Latency: Compiled byte-range AST scanners evaluate commands in < 20 µs (1,000x faster than network filters).
Deterministic Protection: Immune to jailbreak phrasing, prompt injection, and hallucinated bypasses.
Hermetic Workspace Fencing: Guarantees autonomous agents can never modify files or read secrets outside your active repository.
Cryptographic Attestation: Every allowed or denied action is signed with Ed25519 into an RFC 8785 Merkle audit log.

What Bartholomew Guard Protects Against

Threat Vector	What Rogue Agents Try	Bartholomew Guard Action (<20µs)
Catastrophic Shell Deletions	`rm -rf /`, `rm -rf ~`, recursive purge	VETO: Intercepted before OS boundary
Credential Exfiltration	Exfiltrating `.env`, `id_rsa`, `credentials.json`	MASK: High-entropy keys scrubbed in-flight
Destructive Database Mutations	`DROP TABLE users;`, `TRUNCATE`	VETO: Destructive DDL halted
Insecure Privilege Escalation	`sudo`, `chmod 777 /`, `chown root`	VETO: Privilege escalation vetoed
Model Context Leakage	Leaking Stripe, OpenAI, AWS, or GitHub tokens	REDACT: Instant in-memory zeroization

Multi-Assistant Support

Bartholomew Guard provides native 1-click protection across your entire development stack:

Cursor IDE (Rule immunization via .cursorrules & .cursor/rules/)
Claude Code CLI (In-process terminal wrapper & tool execution sandbox)
Windsurf AI (Cascade workflow & tool gating)
GitHub Copilot (Pre-commit hook & workspace confinement)
Cline / Roo-Code (MCP security proxy integration)
VS Code / VSCodium (Activity bar telemetry, dashboard & passkeys)

Quickstart

1. Install the Extension

From Cursor, Windsurf, or VS Code:

code --install-extension Bartholomew.bartholomew-guard-vscode

2. 1-Click Link & Arm Workspace

Press Ctrl+Shift+P (or Cmd+Shift+P on macOS) and run:

Bartholomew: Link & Arm Workspace (1-Click)

This automatically:

Configures deterministic AST rules for your project boundary.
Sets up the fail-closed pre-commit sentinel hook.
Activates in-process secret scrubbing and Keystone spend governance.

3. Inspect Live Telemetry

Click the Bartholomew Shield icon in your activity bar to view your live execution telemetry, active passkey clearances, and cryptographic audit receipts.

Enterprise Compliance & Auditing

EU AI Act Conformity: Article 14 (Human Oversight) and Article 15 (Cybersecurity & Accuracy) attestations.
SOC 2 Type II Evidence: RFC 8785 Merkle receipts generated for CC6.1, CC6.6, and CC7.1 controls.
Air-Gapped SCIF Support: 100% offline mathematical verification with zero external egress.

Community & Feedback

Official Portal: https://bartholomew.info
GitHub Repository: https://github.com/ivegotahunnitonit/bartholomew
Report Issues / Feedback: https://github.com/ivegotahunnitonit/bartholomew/issues

If Bartholomew Guard saved your workspace from an accidental deletion or credential leak, please leave a review on the marketplace and star the repository!

Bartholomew Guard — AI Agent Firewall, Guardrails & Security for Cursor, Claude & Copilot

Bartholomew