AI agents are running code on your machine. Who said stop?
When Claude Code reads your secrets or Cursor installs a rogue package, Guard steps in before the tool call completes.
Free account · Welcome email with setup guide · No credit card required
The problem Guard solves
AI agents are fast.
Blind spots are faster.
The same capabilities that make AI coding agents useful make them dangerous without a control layer between them and your system.
Without Guard
Does your agent know what it just installed?
npm packages from AI-generated install commands run immediately with no receipt, no intercept, no record.
Who approved that MCP server registration?
Remote MCP endpoints extend what your harness can call. Without Guard, they register silently.
Is your .env still yours?
Tool calls like Read(.env) happen in a single round-trip. By the time you see it in logs, it already ran.
With Guard
Every install is wrapped
hol-guard protect wraps npm, pip, and cargo installs. You see what ran, what it found, and the decision it made — every single time.
MCP registration paused
Remote endpoint registrations stop before execution. Guard shows you what the harness is trying to add and waits for your approval.
Secret-bearing file reads intercepted
Pre-tool hooks fire before .env, .npmrc, and key files are opened. You approve or deny before the round-trip completes.
Free forever · No account required · Works offline
Security you can verify, not just trust
Local-first, always
Your code, prompts, configs, and receipts stay on your machine unless you explicitly opt in to cloud sync. Guard never reads file contents server-side.
Exportable receipts
Every block, review, and approval is captured in a local decision receipt you can inspect, export, and sync across devices via Guard Cloud.
No lock-in
Open policy format, exportable receipts, Apache-licensed CLI. If Guard disappears tomorrow your data and your decisions remain yours.
Why developers trust it
Built without blind trust by default.
Guard builds trust by being auditable, local-first, and fully open.
Open source core
The full Guard runtime is published under MIT. Read every line before you run it.
Local-first by design
Scans, approvals, and decisions stay on your machine until you explicitly opt in to cloud sync.
Built for teams
Shared policy packs and investigation routing mean your whole team reviews once, not individually.
For maintainers and plugin publishers
For maintainers
The scanner
for agent
ecosystems.
Validate extensions before release. Use plugin-scanner for maintainer CI checks, then use hol-guard locally to enforce runtime decisions.
# Local maintainer gate
$ pipx install plugin-scanner
$ plugin-scanner verify .
# GitHub Actions PR gate
permissions:
contents: read
security-events: write
jobs:
scan-plugin:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- uses: hashgraph-online/ai-plugin-scanner-action@v1
with:
plugin_dir: "."
min_score: 80
fail_on_severity: highMulti-ecosystem detection
Auto-detects Codex plugins, Claude Code plugins, Gemini CLI extensions, and OpenCode workspace bundles in any repository.
Trust scores and badges
Computes a weighted trust score across security, MCP posture, installability, and maintenance. Plugins that pass can display Scanner Clean, MCP Hardened, and Marketplace Ready badges.
GitHub Action for CI
Use plugin-scanner verify in CI, or the published ai-plugin-scanner action, to gate PRs before release.
MCP transport hardening
Flags insecure HTTP MCP endpoints, wildcard binds, and missing auth postures before they reach production.
Manifest and marketplace validation
Checks relative paths, required screenshots, privacy policy URLs, and plugin metadata completeness — the same rules the HOL Registry uses for import.
Skill-level security scanning
Scans bundled skills for prompt injection markers, zero-width characters, dangerous shell commands, and secret leaks.
Your agent
just got teeth.
Install Guard locally in 30 seconds. Intercepts run on your machine. No account required to start.
Pick your harness
hol-guard install codexhol-guard install claude-codehol-guard install cursorhol-guard detect geminihol-guard detect opencode