HOL Guard: Shield Your AI Harness Before Tools Run

Where this fits

You've just installed a promising Codex plugin or MCP server from a marketplace, and your agent is about to execute it against your local environment. Without a security layer, that tool could read files, exfiltrate data, or mutate your config before you notice. HOL Guard sits between your harness and the tools it runs, acting as an antivirus and approval gate for everything your agent loads.

Why agents benefit

• Pre-flight scanning: The plugin-scanner CLI checks plugins, skills, and MCP servers for known risky patterns before they ever execute.
• Approval receipts: Every guarded action produces a signed receipt, giving you an audit trail of what your agent was allowed to do.
• Reversible protection: Instead of manually editing harness config files, you use Guard commands that can be rolled back cleanly.
• Independent CLI checks: hol-guard and plugin-scanner are separate binaries, so you can verify each one's presence without coupling.
• Release gates: Before shipping a skill or package, run the scanner to catch issues that would otherwise surface in production.

Practical scenarios

Protecting Claude Code from a rogue MCP server

You add a new MCP server for file search. The server's config includes a post-install hook that could overwrite your shell profile. HOL Guard flags the hook during scanning and blocks the server from starting until you explicitly approve it.

Auditing a Codex plugin before first use

A teammate shares a plugin that claims to automate git workflows. You run hol-guard review and see the exact files and commands it will access. You approve only the read-only operations, and the plugin runs with reduced privileges.

Verifying a marketplace package before release

You're publishing a skill to a public marketplace. Running plugin-scanner verify catches a hidden dependency on a deprecated API and a missing license header. You fix both before the package goes live, avoiding negative reviews and security warnings.

Add it to your agent workflow

Install the skill with the OpenAgentSkill CLI:

npx skills add hashgraph-online/hol-guard-plugin --skill hol-guard

Then check that both tools exist:

command -v hol-guard && command -v plugin-scanner

If you're setting up runtime protection, run hol-guard init and follow the prompts. For a quick scan of your current setup:

hol-guard scan --scope all

For a deeper dive, see the skill page on OpenAgentSkill for usage patterns and edge cases.

Compare before adopting

Before committing, compare HOL Guard with alternatives like gitleaks (secret scanning only) or semgrep (static analysis). Evaluate maintenance freshness: check the GitHub repo's commit history and issue response times. Also consider workflow fit—does it integrate with your existing harness configuration, or does it require manual steps you'll forget? Finally, test the approval flow with a dummy plugin to ensure the friction is acceptable for your team.

Why it is worth tracking

HOL Guard addresses a growing gap in agent security: runtime protection for locally executed tools. With only 4 stars, it's early, but the author is active and the Apache-2.0 license invites contribution. The skill's hard rules—like never reading .env files and preferring reversible commands—show a thoughtful security posture. Track it now to evaluate whether it matures into a standard layer for agent harnesses; if the project gains community traction, it could become a default part of secure agent setups. Revisit quarterly to check for new scanning capabilities and integration options.