AI Agent Skills

nsyte publishes an Agent Skill — a SKILL.md document that AI coding agents load on demand. With it installed, an agent knows how to install nsyte, write a valid .nsite/config.json, authenticate with a NIP-46 bunker or --sec, preview deploys with --dry-run, deploy to Blossom servers, inspect and update published sites, set up CI/CD, and work through common errors — without you pasting docs into the conversation.

The skill follows the open Agent Skills format and works with Claude Code, Codex, Cursor, OpenCode, GitHub Copilot, Gemini CLI, Cline, Windsurf, and every other agent the skills CLI supports.

Install

npx skills add sandwichfarm/nsyte

The CLI detects which agents you have installed and links the skill into each one's skills directory. Useful options:

Option	Effect
-g, --global	Install for your user (all projects) instead of this project
-a <agent>, --agent	Target specific agents, e.g. -a claude-code -a codex
-y, --yes	Skip confirmation prompts (scripts / CI)
--copy	Copy files instead of symlinking
--skill nsyte	Select the skill explicitly (the repo currently ships one)

Later:

npx skills list            # show installed skills
npx skills update          # pull the latest version of every installed skill
npx skills remove nsyte    # uninstall

Manual install

The skill is a plain directory — copy skills/nsyte/ from the repository into your agent's skills folder:

git clone --depth 1 https://github.com/sandwichfarm/nsyte /tmp/nsyte

# Claude Code, this project
mkdir -p .claude/skills && cp -r /tmp/nsyte/skills/nsyte .claude/skills/

# Claude Code, user-wide
mkdir -p ~/.claude/skills && cp -r /tmp/nsyte/skills/nsyte ~/.claude/skills/

# Agent-neutral directory read by many agents
mkdir -p .agents/skills && cp -r /tmp/nsyte/skills/nsyte .agents/skills/

What the skill contains

skills/nsyte/
├── SKILL.md                  # instructions: setup, config, auth, deploy, inspect, CI, troubleshooting
├── references/
│   └── nostr-concepts.md     # relays, pubkeys, nsec, Blossom, NIP-46, event kinds — for agents new to Nostr
└── assets/
    └── config.schema.json    # JSON Schema for .nsite/config.json

The skill is deliberately conservative: it tells agents to prefer --dry-run before publishing, never to print or commit secrets, never to hand-edit bunkerPubkey, and to confirm with you before running delete or undeploy.

Keeping it current

The skill is checked in CI by the same drift gate that validates the command reference pages (deno task check-doc-drift): every command must be mentioned, every flag the skill references must exist in source, and the bundled config schema must match src/schemas/config.schema.json. If you notice the skill giving stale advice, please open an issue or a PR against skills/nsyte/.

See Also

• Installation — install the nsyte CLI itself
• CI/CD — the automation flow the skill teaches agents
• Security — how credentials are stored