up

name

up

description

Spin up a fresh dev environment in one shot — create a new git worktree off the latest master, provision it with the regular `pnpm worktree:up` (a private CLOUD dev deployment + seed), then start BOTH dev servers (Next + Convex watcher) as background tasks and verify the app responds. Use when the user says "/up", "spin up dev", "spin up a worktree", "new worktree", "start a fresh worktree", or otherwise wants a ready-to-use dev environment at the state of master. Dev-only — never touches production.

Up

Get from "I want to work on something" to "a dev server is serving the app at the state of master" in one guided flow: new worktree off fresh master → pnpm worktree:up → start both servers → verify.

This is the regular (default) worktree:up path — a private CLOUD dev deployment. Do NOT use --local / plane mode here. (Plane mode is a separate, explicit choice; see "Local Convex backend" in CLAUDE.md.)

Everything here is dev-only and non-destructive, so run all steps automatically without approval gates.

🚫 Safety rules (read first)

• Never touch production (fantastic-dotterel-572): no convex deploy, no prod env changes. This skill only creates a dev worktree + dev deployment.
• Branch off fresh origin/master, not whatever the current checkout happens to be — git fetch first so the worktree reflects the latest master.
• Don't reuse port 1041 (master checkout) — worktree:up assigns this worktree its own port automatically; never override it to 1041.
• Start servers, don't clobber existing ones. Each worktree has its own port + private deployment, so a new one won't collide; just don't kill another checkout's servers.

Set the toolchain PATH once at the top of every command: export PATH="$HOME/.local/share/mise/shims:$PATH" (mise-provided node/pnpm).

Step 1 — Create the worktree off fresh master

git worktree add must target a path under the master checkout (compute it so this works whether you're standing in master or another worktree). Pick a short, memorable Docker-style name (adjective-surname); append a random hex suffix so it never collides. Optionally seed the adjective/surname from what the user said they'll work on.

export PATH="$HOME/.local/share/mise/shims:$PATH"
MASTER="$(dirname "$(git rev-parse --path-format=absolute --git-common-dir)")"
cd "$MASTER"
git fetch origin master --quiet
NAME="lucid-mclean-$(openssl rand -hex 3)"      # ← pick your own adjective-surname
git worktree add -b "claude/$NAME" ".claude/worktrees/$NAME" origin/master

The branch (claude/<name>) tracks origin/master, so it starts exactly at the latest merged commit.

Step 2 — Provision it (private cloud dev deployment + seed)

Run the regular pnpm worktree:up from INSIDE the new worktree. It is idempotent and does the whole setup: symlinks node_modules, copies .env.local, provisions + selects a private cloud dev deployment (so it never shares master's perceptive-husky-735), assigns + persists a unique port, points origin-bound auth (PASSKEY_ORIGIN/SITE_URL) at it, enables /dev-login, pushes schema + functions, seeds test data + standards, and writes .claude/launch.json. Takes a few minutes (it pushes + seeds).

cd "$MASTER/.claude/worktrees/$NAME"
pnpm worktree:up

It finishes with ✅ Worktree ready on its own dev deployment, port <PORT>. — note that port (also persisted as PORT= in this worktree's .env.local).

Step 3 — Start both dev servers as background tasks

Start BOTH as Claude background Bash tasks (not detached, not a separate terminal) so they appear in the task list and a crash re-pings you. Use the worktree dir as cwd and set the mise PATH in each.

# Background task A — Next dev server (reads PORT from .env.local, --webpack):
pnpm dev:worktree

# Background task B — Convex function watcher (headless once provisioned):
npx convex dev

Step 4 — Verify the app responds

Don't claim success until the port actually answers. Read the assigned port from .env.local and curl the dev-login URL:

PORT="$(sed -n 's/^PORT=//p' .env.local)"
curl -s -o /dev/null -w "HTTP %{http_code}\n" "http://localhost:$PORT/dev-login?u=avery&to=/teacher"

Expect HTTP 200. Also confirm the Convex watcher logged Convex functions ready! (queries fail until it has). If the Next log shows it compiled /dev-login and returned 200, you're up.

Finish

Report:

• the worktree path (.claude/worktrees/<name>) and branch (claude/<name>),
• the master commit it's based on,
• the port and the ready-to-click sign-in URLs, e.g.
  • http://localhost:<PORT>/dev-login?u=avery&to=/teacher (admin)
  • http://localhost:<PORT>/dev-login?u=test-teacher-001&to=/teacher
  • http://localhost:<PORT>/dev-login?u=test-scholar-001&to=/scholar
• that both servers are running as background tasks.

When the session's done, the cleanup skill tears this down (stops the servers, optionally removes the worktree + branch).