# Quickstart Five steps take you from nothing to a team with an evaluated, published skill. Each is a command or two, with a pointer to the page that covers it in full. ## What you need git and Node.js 22.12 or later. To create a team on GitHub, a logged-in GitHub CLI (`gh auth login`); joining a team needs none. To run evals, Claude Code installed and logged in. [Install](https://www.terum.ai/docs/install) has the full list. ::: warning Every eval runs real Claude Code sessions billed to your own account, and a single `eval` shows no estimate before it starts. ::: ## The five steps ::: steps ### Run setup Create a team with bare `setup`, or join one by naming its repository. ```sh npx -y terum-skills@latest setup npx -y terum-skills@latest setup / ``` Creating makes a private GitHub repository under your account, records you as its first member, and invites the teammates you name. Joining accepts your invitation, clones the team repository and prints `Joined as `. On macOS and Windows, setup opens the [desktop app](https://www.terum.ai/docs/desktop-app) and continues there. See [Create a team](https://www.terum.ai/docs/create-a-team) and [Join a team](https://www.terum.ai/docs/join-a-team). ### Find and install a team skill `ls` lists the members and shared skills, and `search` matches a term against each skill's name, description and category. Both read your local clone. ```sh npx -y terum-skills@latest search npx -y terum-skills@latest install ``` `install` copies the latest version into Global (`~/.claude/skills`) or into a project you registered, and asks which. Joining installs nothing on its own. A team you just created has no skills yet, so go on to step 3 with one of your own. See [Install and manage](https://www.terum.ai/docs/install-and-manage). ### Evaluate a skill Pick any folder in your Library: the skill you just installed, or one of yours. ```sh npx -y terum-skills@latest eval ``` Missing test cases and trigger prompts are generated into the folder first. Each case then runs in fresh sandboxes: without the skill (the baseline), with your folder (the candidate), and with the team's most recently evaluated version (the incumbent) when there is one. The verdict is banded on net lift against the baseline: `PASS` at or above +1/3, `FAIL` at or below -1/3, `NEUTRAL` between. A run costs cases × k × arms sessions, with `k` defaulting to 1; `--triggers-only` is the cheapest look. When the bytes are already a published version, as an unedited installed copy's are, the receipt goes straight to the team. See [How evals work](https://www.terum.ai/docs/evals). ### Publish one of your own skills The folder must be in Global or a project you registered. ```sh npx -y terum-skills@latest publish ``` Publish adds the team's managed fields to your `SKILL.md`, runs the [hygiene checks](https://www.terum.ai/docs/hygiene), and commits the folder as the next immutable version, `skills//v/`. It attaches any local receipt for exactly those bytes, and asks first if the newest one is a `FAIL`. A first publish prints `Published deploy-check as Version 1 to the acme marketplace. Attached 0 eval run(s).` Your folder now holds the published bytes, so an `eval` of it afterwards shares its receipt with the team. See [Publish a skill](https://www.terum.ai/docs/publish). ### Stay current Reads never fetch, so a teammate's publish reaches you when something runs `sync`. ```sh npx -y terum-skills@latest sync ``` Sync fetches the team repository and places nothing. The session hook, if you accepted it, fetches when a Claude Code session starts, at most once an hour; the desktop app fetches when its window regains focus, at most once a minute. To take a newer version of a skill you installed, run `install` again. See [Claude Code integration](https://www.terum.ai/docs/claude-code). ::: ## Where to go next ::: cards - [How it works](https://www.terum.ai/docs/how-it-works): The model, what runs where, and why it is built this way. - [Library and Marketplace](https://www.terum.ai/docs/library-and-marketplace): Your folders on disk versus the team's published versions. - [Results and receipts](https://www.terum.ai/docs/results): Every line of the eval report and what a receipt records. - [Running evals](https://www.terum.ai/docs/running-evals): Batches, the queue, overnight runs and every flag. :::