Zed, told what it is about to install

What one command wires

Wire Zed to Vulnetix. Zed reads the interoperable ~/.agents/skills, and its context-server configuration points at the MCP server.

vulnetix agent install --agent zed

• Agent Skills: supported — ~/.agents/skills
• Dependency guard: not wired by the installer — hook contract not yet verified
• MCP server: supported — https://mcp.vulnetix.com/mcp

Where Zed reads skills

Agent Skills is an open standard, but hosts disagree about the directory. The installer writes the one Zed actually opens.

• Everywhere, for you: ~/.agents/skills
• This repository: .agents/skills

Add the MCP server to settings.json

{
  "context_servers": {
    "vulnetix": {
      "source": "custom",
      "url": "https://mcp.vulnetix.com/mcp",
      "headers": { "Authorization": "ApiKey <orgId>:<key>" }
    }
  }
}

Ask something training data cannot answer

No install and no local binary: one URL and one header give the agent live vulnerability intelligence it can call mid-conversation.

>
 is CVE-2021-44228 still actively exploited?
→ calling vulnetix_vuln
org.apache.logging.log4j:log4j-core@2.14.1 · maven
17 vulnerabilities · 11 critical, 4 high
- CVE-2021-44228  critical · CVSS 10.0 · EPSS 100.0%
- CVE-2021-45046  critical · CVSS 9.2  · EPSS 100.0%
- CVE-2021-44832  critical · CVSS 9.3  · EPSS 97.9%
CISA KEV · due 2021-12-24 · known ransomware use

One stage of four

• Build — agents, guarded at the tool call
• Observe — AI-BOM finds what the repository declares
• Govern — allow and deny tools, MCP servers and skills
• Prove — EU AI Act, NIST AI RMF and ISO 42001 evidence