All projects

AI agent skill · Network operations

VPNOpsNyx

A skill that teaches Claude, Codex and other AI agents to inspect VPN panels, nodes, tunnels, relays and DNS, test paths with real data, and wait for the operator’s approval before any change.

Public source codeFrom the project archive
Original project illustration, not a screenshot of the product.

The problem

A VPN fleet fails in ways a status page never shows. A relay answers 200 and then stalls after 15 KB, a tunnel replies to ping and carries nothing, and a datacenter rotates an IP so that every mode that worked yesterday is gone.

What the product does

For operators who run a VPN or proxy service, the skill gives an agent the routine of an experienced operator: identify the target, read the matching runbook, stay read-only until a change is approved, verify by real data flow, and report what changed and how to roll it back. It carries runbooks for six panels and eight tunnel families, with references on relays, Cloudflare DNS, capacity and filtering in Iran.

Behind the technology

It is not a VPN, a tunnel or a panel: it is Markdown, JSON and small helper scripts that an agent loads through SKILL.md, which routes it to the right runbook. A path is judged by the data that actually arrives, tested from a probe that is not the relay itself, and a new path is checked again the next day. One helper writes nftables DNAT rules, validates them with nftables and offers a dry run; the agent is told never to print or store credentials, keys or subscription links.

  • SKILL.md
  • Markdown
  • JSON
  • Shell
  • Python
  • nftables