AI agent setup

If you build with an AI coding agent, such as Claude Code, Cursor, Codex, Copilot or Windsurf, it can do the setup part of From zero to your module in your sandbox for you. Paste this one line into your agent:

Fetch https://developers.tango.vision/agent-setup/prompt.md and follow the instructions.

Before you start

• Your registry token in TV_NPM_TOKEN. It arrives in a separate message from your Tango Vision contact. Export it in your shell profile or secret manager before you start the agent.
• For the sandbox part, your portal API key in TV_API_TOKEN, or a session from npx @tv/extension-sdk login. Without either, the agent sets up the module and leaves the sandbox to you. See Get a sandbox API key.

What the agent does

1. Checks Node 24 (22 also works) and pnpm 10.18.2. It asks you before changing your global pnpm.
2. Adds the @tv registry to your ~/.npmrc in the form that reads the token from TV_NPM_TOKEN, so the file holds no secret, and checks that @tv packages resolve from that registry, not from the public npm registry.
3. Asks for the module's slug, your company's npm scope and name, and a category, then scaffolds with init-module --external --scope --author and runs pnpm install.
4. Runs typecheck, test, build, validate and check-exposes.
5. Keeps the AGENTS.md the scaffold writes and adds only a pointer to these docs. It follows the rules on Calling the API from a module: useSelectedBuildingId() for the building, authorizationHeader() on every request to a relative /api/... path, and usePlatformContext() in tests only, because it throws inside Building OS.
6. Lists your sandboxes and asks you what to do: use one, wait for access to the one a colleague created, create a new one, or skip. It creates a sandbox only when you say so, because your team may share one (Two developers, one sandbox).
7. Stops at the hand-off. It asks who else works on the sandbox and gives you the message to send your contact, with the sandbox's slug and those names, and a report of what is left for you: serving the module, installing it into the sandbox, and signing in.

What the agent does not do

• Accounts and credentials. It does not create your developer account, sign in for you or mint keys, and it never asks you to paste a token into the conversation.
• Sandboxes without your yes. It never creates, resets, extends or deletes a sandbox unless you agreed to that exact action.
• The hand-off. You send the sandbox's slug to your contact yourself and receive your sandbox login.
• Anything after the hand-off, unless you ask it to continue. It then follows the walkthrough from step 4.

The prompt itself

The instructions your agent fetches are plain Markdown at a stable URL, so you and the agent can read them at any time. They open in the browser as text:

→ /agent-setup/prompt.md

They work with @tv/extension-sdk 1.15.0 and later.