graiz/local.vibe/vibeskill

local.vibe

local.vibe (CLI: vibe) is a local DNS daemon, already installed and running on this machine, that gives every dev server a friendly, collision-free https://<name>.vibe address.

The rule: never guess a port or a localhost URL

When you start or reference a local server, do not pick a port yourself, write http://localhost:3000, or assume the framework's default. Two projects that both grab 3000 clobber each other. Let vibe assign and track the port, then use the https://<name>.vibe URL it hands back.

Starting a server (the two commands you need)

• In a project that has a `vibe.json`: vibe start
• Ad hoc, no config file: vibe start <name> <port> -- <command>

Pass 0 for the port to let vibe auto-assign one and inject it as $PORT. Do not use this form inside a git worktree — see below.

Minimal vibe.json (omit port so vibe assigns one and sets $PORT):

{ "name": "myapp", "cmd": "npm run dev" }

The command must bind $PORT (read process.env.PORT / os.environ["PORT"], or pass --port $PORT). The app is then reachable at https://myapp.vibe.

Git worktrees — run vibe start, change nothing else

Inside a linked worktree, the whole procedure is:

vibe start            # from the worktree directory. That's it.

You get https://<branch-slug>.<app>.vibe on its own auto-assigned port. The main checkout's <app>.vibe and every other worktree keep running untouched — vibe already guarantees that isolation, so you never need to arrange it yourself.

Do not "avoid a collision" — there isn't one. Specifically, in a worktree:

• Do not change `name` in the worktree's `vibe.json`. It is a copy, and

renaming it (myapp → myapp-wt) is the single most common way to break this. The worktree stops being a worktree of the app: it disappears from the app's picker and dashboard group, and it no longer inherits the app's oauth_callback_port, so OAuth logins break. vibe reads the app name from the main checkout and will tell you when the copy disagrees — but leave the copy alone.

• Do not register a second app with vibe start <name> <port> -- <cmd>.

That creates an unrelated route, not a worktree of the app.

• Do not pick a port, and do not copy the parent's port — vibe assigns a

free one and injects $PORT.

Only the subdomain is yours to choose, via vibe start --as <name> if the branch slug is unwieldy.

A worktree inherits the parent app's OAuth callback bridge, and vibe copies the parent's untracked .env* files in if the worktree has none. Deleting the worktree removes the route automatically.

Finding what's already running (don't guess the URL)

• vibe list — every route and its https://<name>.vibe URL
• vibe status — whether the daemon is up
• vibe doctor — diagnose and repair the request path

Full guide

For framework-specific config (Vite/Next allowedHosts, Flask, Django, Rails, Jekyll), auto-port details, reserve_ports for multi-port apps, worktree routes, and OAuth localhost callbacks, read the authoritative guide from the running daemon — don't reproduce it from memory:

curl -sS --max-time 5 http://localhost:7999/setup.md || curl -sS --max-time 5 http://local.vibe/setup.md

The daemon's fixed port is tried first so the guide is reachable even when DNS or the privileged-port redirect is down. The local.vibe fallback covers the opposite case: a VPN kill-switch or firewall that filters direct connections to loopback high ports while the redirect still works. If both fail, run vibe doctor.