Skip to content

Claude Code skills

SPIRENS ships a Claude Code plugin (spirens-skills) that teaches an AI agent the decentralized-web infrastructure this repo deploys — so it can help you stand up the same stack on your own box, not just read the docs.

Modeled on ethskills. ethskills covers Ethereum itself; spirens-skills covers the infra that makes a decentralized-web deployment actually serve traffic.

Install

# Inside a Claude Code session:
/plugin marketplace add MysticRyuujin/spirens
/plugin install spirens-skills

After install, Claude routes questions like "how do I get a wildcard cert for *.ipfs.example.com?" to the right skill on demand.

What's in the plugin

Each skill is a short, triggerable markdown file. Corrections, not tutorials — they exist to patch specific LLM blind spots.

Start here

  • ship — end-to-end path from bare domain to live https://rpc.example.com. Fetch FIRST; routes to the rest.

Naming & discovery

  • dns — records, TTL, CAA, propagation reality.
  • cloudflare — proxy vs DNS-only, API tokens, Universal SSL interaction.
  • ipns — keys, publishing, TTL, and why dnslink is usually what you want.
  • ens-resolution — ENS name → contenthash → IPFS CID → gateway.

Transport & certs

  • tls-acme — ACME protocol, DNS-01 vs HTTP-01, wildcards, rate limits.
  • lets-encrypt — LE specifics: chain of trust, renewals, ECDSA vs RSA.

Content routing

  • ipfs — Kubo, pinning, peering, DHT vs delegated routing.
  • gateways — subdomain vs path, trusted vs trustless, origin isolation.

Reverse proxies

  • traefik — routers/services/middlewares, ACME, file vs label provider.
  • nginx — http/stream, TLS termination vs passthrough, certbot for ACME.
  • caddy — automatic HTTPS, Caddyfile, on-demand TLS risks.

JSON-RPC

  • erpc — finality-aware caching, hedging, failover, per-chain config tiers.
  • helios — a16z's trustless light client; eth_getProof requirement, checkpoint trust, placement between dweb-proxy and eRPC.

Topology

  • topology — single-host Compose vs Docker Swarm, overlay networks, stateful service constraints.

How skills work with the existing docs

Skills are short, link-heavy and designed to be triggered in an AI conversation. The canonical docs/ tree (this site) is long-form and designed to be read top-to-bottom.

A skill will:

  1. Open with a "What You Probably Got Wrong" section correcting common LLM misconceptions.
  2. Give the minimum concrete how-to.
  3. Link into docs/ for the full walkthrough and into config/ / compose/ for the actual working config SPIRENS ships.

If you're reading the docs, you usually don't need the skills. If you want an AI pair to help you deploy this somewhere that isn't exactly SPIRENS, the skills are the portable, triggerable layer.

Source

The plugin manifest is at .claude-plugin/plugin.json and .claude-plugin/marketplace.json. Each skill is a directory at the repo root containing a SKILL.md.