Openclaw Security Guard

Security audit CLI + live dashboard for OpenClaw. Scans for secrets, config issues, prompt injections, vulnerable dependencies, and unverified MCP servers. Zero telemetry.

Security Scan

The skill's code, install method, and runtime instructions are consistent with a Node.js-based OpenClaw security scanner — nothing obviously disproportionate or covert, but a few small metadata mismatches and the usual need to verify the 'zero telemetry' claim mean you should review the package before trusting it with real systems.

Purpose & Capability

Name/description match the delivered pieces: a Node.js npm package that exposes CLI binaries and implements secrets/config/prompt-injection/dependency/MCP-server scanners and a local dashboard. Required binary (node) and the npm install are proportionate to the stated purpose. The publish metadata omission of a source/homepage in the registry (but SKILL.md includes a GitHub URL) is a small inconsistency worth verifying.

Instruction Scope

SKILL.md instructs the tool to scan the user's OpenClaw install (default paths like ~/.openclaw), run an optional auto-fix that edits configuration, and open a localhost dashboard. Those actions are expected for a security auditor. Note: auto-fix modifies user files (claims to backup first) — this is expected but the user should confirm backups and review proposed fixes before running --auto. The docs include example malicious prompt strings (used to demonstrate the prompt-injection detector); that's why prompt-injection patterns appear in the docs.

Install Mechanism

Install uses an npm package (openclaw-security-guard) which is appropriate for a Node.js CLI. npm install is a standard distribution method; npm packages carry typical supply-chain risk, so verify package provenance and version before installing globally.

Credentials

The skill declares no required environment variables or credentials. Documentation references optional env vars (OPENCLAW_HOME, OPENCLAW_GUARD_CONFIG), which is reasonable for a local scanner. No unexplained credential or system-wide config access is requested in metadata or SKILL.md.

Persistence & Privilege

always is false and model invocation is default — normal. The package runs as a CLI and dashboard service and can modify OpenClaw config when asked (auto-hardening). It does not request unwarranted system-wide privileges or attempt to persist across unrelated skills. Verify where the dashboard auth file is stored (docs mention ~/.openclaw-security-guard/auth.json) if you are concerned about local persistence.

Scan Findings in Context

[ignore-previous-instructions] expected: The SKILL.md and docs include example prompt-injection test strings (e.g., 'ignore previous instructions') as part of the Prompt Injection Detector documentation and programmatic examples. The detector flag is expected for a tool that demonstrates such patterns, but you should still review the code for any runtime behavior that executes or forwards untrusted content.

Assessment

This package appears coherent for a local OpenClaw security scanner, but take these precautionary steps before installing or running with --auto: - Verify the npm package and repository: confirm the package on npm matches the GitHub repo referenced in SKILL.md and check the package author/publisher identity. - Grep the source for outbound network calls (http, https, ws, fetch, axios, net.connect) to confirm 'zero telemetry' — focus on dashboard/server.js, monitors/*, and helpers for any external endpoints or hosts. - Review code paths that modify configuration (auto-hardener, fix command) and test fix --dry-run first; ensure backups are created in a location you control. - Inspect where dashboard credentials are stored (~/.openclaw-security-guard/auth.json is mentioned) and secure or delete that file as needed. - Prefer running via npx or in an isolated/sandbox environment initially rather than global install. If you want, I can (1) point to specific files to grep for outbound connections or secrets exfiltration patterns, or (2) run a quick static checklist of the top files (dashboard/server.js, monitors/*, auto-hardener.js) and list lines that look like external network usage. Confidence is medium because the package includes full source (which helps) but registry/source metadata had a small mismatch and claims like 'zero telemetry' should be verified by code inspection.

The missing security layer for your OpenClaw installation.

What it does

Run openclaw-guard audit to scan your OpenClaw setup across 5 categories:

• Secrets Scanner -- Detects API keys, tokens, passwords across 15+ formats + entropy analysis
• Config Auditor -- Checks sandbox mode, DM policy, gateway binding, rate limiting
• Prompt Injection Detector -- 50+ patterns: instruction overrides, role hijacking, jailbreaks
• Dependency Scanner -- npm CVE scanning
• MCP Server Auditor -- Allowlist-based verification of installed MCP servers

Quick start

npm install -g openclaw-security-guard

# Full audit
openclaw-guard audit

# Fix issues automatically (with backup)
openclaw-guard fix --auto

# Launch live dashboard
openclaw-guard dashboard

Features

• Security Score (0-100) -- one number for your security posture
• Auto-hardening -- interactive, automatic, or dry-run modes
• Live dashboard -- real-time monitoring at localhost:18790
• Pre-commit hooks -- catch secrets before they're committed
• Multi-language -- English, French, Arabic
• Zero telemetry -- no tracking, no network requests, 100% local

Links

• Repository: https://github.com/2pidata/openclaw-security-guard
• Author: Miloud Belarebia / 2PiData
• License: MIT