Skills

Runnable capabilities, grounded in Norien's live data. Publish a skill once; anyone can run it and get a real, data-backed result.

A skill pairs a plain-language playbook with a Norien data source. When you run it, the backend resolves that data — a wallet's portfolio, the market list, one token, or the registry — and hands it to a fast model as grounded context. The output is about real state, not an ungrounded guess.

Where a tool is a single function and an agent is a whole program, a skill sits between: one packaged capability you invoke on demand, from the app or the CLI.

Starter skills

market-recap — a crisp summary of today's Robinhood Chain movers.
review-wallet — analyzes any wallet's multi-chain portfolio.
explain-token — explains a token from its live market data.
find-agent — recommends agents and tools for what you want to do.
draft-agent — turns a description into a valid agent.json.

Run a skill

From the app, open Skills, pick one, and run it in the browser. From the terminal:

norien skill run market-recap
norien skill run review-wallet 0xd8dA6BF26964aF9D7eEd9e03E53415D37aA96045
norien skill run explain-token <token-address>

Or over the API — it streams the result as server-sent events:

curl -N -X POST https://api.norien.live/api/skills/market-recap/run \
  -H "Content-Type: application/json" \
  -d '{ "input": "top movers today" }'

Publish your own

Write a skill.json and publish it — it becomes runnable by anyone, no code to deploy. Pick a data_source and the platform grounds every run for you.

{
  "slug": "gas-check",
  "name": "Gas Check",
  "description": "Explains whether now is a cheap time to transact.",
  "instructions": "From the market data, comment on activity and suggest whether conditions look calm or busy. Keep it to two sentences.",
  "data_source": "markets",
  "input_hint": "optional focus",
  "examples": ["is it busy right now?"]
}

norien skill publish

Data sources

markets — live top gainers and highest-volume tokens.
portfolio — the input is a wallet address; its holdings are resolved.
token — the input is a token address; its market data is resolved.
registry — searches agents and tools for the input.
none — a pure instruction skill, no data resolved.

Note —A skill's data source is a fixed, safe set — never arbitrary code — so running someone else's skill is as safe as reading their instructions. Answers are grounded in the data provided; they are not financial advice.