A setup that works for you probably works for someone else too. To share one, add a frontier.json file to a public GitHub repo. It describes your workspaces, your agents with their prompts or scripts, and any Domains they read, along with a name, a short description, and what someone needs ready before they start.
Anyone with Frontier can then open frontier://try/github/<owner>/<repo>. Before anything is created they see a review screen with every workspace, every agent's full prompt or script, and every tool a script is allowed to call. Every agent arrives switched off, so nothing runs until they choose to turn it on. See Agents & the Conductor for what each agent trigger means.
Setups we can install are listed on the Examples page, each with its own page you can link to. That page shows your name and X handle, the post you shared it in, and a video of it running when there is one. Share your repo on X and we may find it. A listing always points at the exact commit that was checked, so a later push never changes what people install until it has been checked again.
A repo doesn't need a frontier.json to be listed. Agent folders written for Claude Code, Codex, GitHub Copilot, Gemini CLI, Cursor, OpenCode or Kiro, and CrewAI crews, are turned into a Frontier setup once, when they're listed: the agent that coordinates the others leads from a parent workspace, each specialist gets a workspace of its own, anything described as recurring becomes a scheduled agent, and every prompt stays the author's own words. The review screen says which tool it came from.
A setup can say what it needs before its agents can work: a command-line tool, a sign-in, an API key, a repository. The listing shows that as a short "Needs: GitHub, Stripe" line, and the review screen lists each one with the exact command it would run. After you install, a "Set up" thread in the new workspace works through the list with you. It checks what's already in place, asks for your approval before it installs or signs in to anything, sends you to Settings → Credentials for keys, and ends by telling you what's left. Anything Frontier already covers, like a browser, is skipped.
A repo that is mostly skills installs as a skills library instead of a team: one lead agent that knows every skill in the pack and picks the right one for each request, with each skill kept as a macro you can type into any message or as reference notes on the workspace.
If you're writing a frontier.json by hand, the full reference with a complete example is the community-setups page in Frontier's own Help area, inside the app.