AI Command Gatekeeper

AI Command Gatekeeper is a VS Code-family extension (VS Code / Cursor / Antigravity) that acts as a security control layer for shell commands proposed by AI agents such as Kilo, Roo, and Cline.

Key Features

• Receives command requests through command hooks.
• Analyzes risk using deny-lists, allow-lists, obfuscation signals, and pipe-to-shell detection.
• Makes policy-driven decisions: allow | ask | deny.
• Executes approved commands in a controlled terminal with safe cwd handling.
• Writes audit logs to the Output channel and JSONL log file.

Installation

1. Install dependencies:

npm install

2. Build:

npm run compile

3. Package VSIX:

npm run package:vsix

4. Install the VSIX in VS Code/Cursor/Antigravity using Extensions: Install from VSIX....

Public Commands

• aiCommandGatekeeper.submitCommandRequest
• aiCommandGatekeeper.submitCommandText
• aiCommandGatekeeper.evaluateCommandRequest
• aiCommandGatekeeper.runClipboardCommand
• aiCommandGatekeeper.applyKnownAgentSettings
• aiCommandGatekeeper.cline.submitShellCommand
• aiCommandGatekeeper.roo.submitShellCommand
• aiCommandGatekeeper.kilo.submitShellCommand

Policy Configuration

• Default policy file: .ai-command-gatekeeper/policy.json
• Can be configured via VS Code settings: aiCommandGatekeeper.*
• Strict policy example: examples/policy.strict.json

Risk Matrix

• critical: always denied (matched blocked patterns)
• high: default is ask (or allow if autoExecute.highRisk is enabled)
• medium: default is allow
• low: default is allow

Documentation

• Agent integrations: docs/integration-kilo-roo-cline.md
• Security model: docs/security-model.md
• Implementation plan: docs/plans/2026-02-19-ai-command-gatekeeper-vscode-family.md