Panguard AI v0.3.0: Skill Auditor, MCP Tools & Public Release

Panguard AI Is Now Public

Today we are making Panguard AI publicly available. Eight security tools, one install command, zero expertise required. Our mission is simple: give every developer and small business the same protection that Fortune 500 companies pay six figures for.

What Is New in v0.3.0

### Skill Auditor (Beta)

AI agents are everywhere. OpenClaw, AgentSkills, MCP tools -- the ecosystem is exploding. But who audits the skills before you install them? Panguard Skill Auditor runs 5 automated security checks on any SKILL.md file: manifest validation, prompt injection detection, tool poisoning detection, dependency analysis, and permission scope analysis. Run `panguard audit skill ./my-skill` and get a 0-100 risk score in seconds.

### MCP Server: 11 Tools

Control Panguard from Claude, Cursor, or any MCP-compatible AI assistant. The MCP server now exposes 11 tools: scan, scan_code, guard_start, guard_stop, status, alerts, block_ip, generate_report, init, deploy, and the new audit_skill. Start with `panguard serve --mcp`.

### Security Hardening

This release includes multiple security fixes identified during our pre-release audit: Manager API now requires authentication tokens in production, waitlist verification tokens are hashed with SHA-256, the remote scanner blocks private IP addresses to prevent SSRF, and Sigma rule regex compilation includes ReDoS protection.

Get Started

Install Panguard in one command:

bash
curl -fsSL https://get.panguard.ai | bash
panguard scan --quick

Panguard is 100% open source under the MIT license. Unlimited machines, unlimited scans, full AI-powered protection. No paid tiers.

What Is Next

We are working on SOAR playbook automation, log collectors for centralized monitoring, and anomaly scoring improvements. Follow us on GitHub for updates.