HOL Guard

HOL Guard is a local security runtime for AI coding agents. Use this skill when a user wants to put a supported local harness behind Guard before tools execute, inspect a blocked or approval-gated action, produce receipts or audit evidence, or verify an agent package before release. The `hol-guard` CLI owns harness setup and runtime protection; `plugin-scanner` is a separate scanner distribution for plugins, skills, MCP server packages, and mixed agent workspaces. Do not claim a harness is protected until HOL Guard reports that state, and never bypass a Guard approval. Supported local harness targets include Codex, Claude Code, Copilot CLI, Cursor, Gemini CLI, Hermes, OpenClaw, OpenCode, and Antigravity.

## Installation

Install the skill from Agent Skill Exchange with the pinned third-party installer documented by the marketplace:

npm exec --package=skills@1.5.7 -- skills add agentskillexchange/skills --skill hol-guard

Install the runtime in an isolated Python environment:

pipx install hol-guard
hol-guard status
hol-guard detect --json

For package scanning, install the scanner separately:

pipx install plugin-scanner
plugin-scanner verify .

## Protect a local harness

Use Guard-owned setup rather than editing agent configuration by hand:

hol-guard bootstrap
hol-guard install 
hol-guard run  --dry-run
hol-guard run 
hol-guard status

For example, use `codex`, `claude-code`, `cursor`, `gemini`, `openclaw`, or `opencode` as the harness name. If Guard queues work, inspect it before approving or denying:

hol-guard approvals
hol-guard approvals open
hol-guard receipts
hol-guard diff 

For diagnostics and evidence:

hol-guard doctor  --json
hol-guard inventory
hol-guard events

## Scan agent packages

Run scanner mode from the package or workspace root so relevant plugin, skill, MCP, and harness configuration surfaces can be discovered together:

plugin-scanner lint .
plugin-scanner verify . --json

Treat scanner failures as real until inspected. Do not mark a package release-ready or a workspace protected without command output proving it.

## Source

– HOL Guard skill and runtime guidance: https://github.com/hashgraph-online/hol-guard-plugin
– HOL Guard: https://hol.org/guard