HOL Guard

HOL Guard is a security tool that sits between you and your AI coding agents — Claude Code, Codex, Cursor, Gemini CLI, and others — and intercepts dangerous actions before they happen. It watches for things like an agent trying to read your ~/.npmrc file, install a suspicious package, register a malicious MCP server, or modify config files without approval. Think of it as a permission gate for AI agents: when an agent attempts something risky, Guard blocks it and lets you decide what to do.

The tool works by installing a lightweight harness that hooks into your agent's runtime. Once set up, it monitors every command the agent tries to run and checks it against a set of security rules. If something looks off — like a prompt injection attempt, a secret exfiltration attempt, or a connection to an unverified MCP endpoint — Guard stops it and alerts you. It also includes a plugin scanner for maintainers that can validate extensions before release, checking for things like insecure HTTP endpoints, missing auth, and dangerous shell commands in bundled skills. The scanner can run in CI via a GitHub Action, so plugin publishers can gate pull requests automatically.

HOL Guard is most useful for developers and teams who rely on AI coding agents and want to prevent accidents or attacks. If you're using Claude Code, Cursor, Codex, or similar tools in production, Guard adds a safety layer that catches mistakes and malicious actions alike. It's also valuable for plugin and extension publishers who want to validate their code before shipping. The tool is free to start — you create an account, get install instructions by email, and can upgrade to Pro later for team policies, Slack alerts, and multi-machine sync.