Dev Server + Monitor

Start the project's dev server in the background and attach a persistent Monitor so you're notified automatically on any error or warning — without polling.

---

Step 1 — Detect the start command

Inspect the project root (current working directory) in this priority order:

1. package.json — use npm run dev if a "dev" script exists, otherwise npm start if "start" exists, otherwise npm run serve if "serve" exists.
2. pyproject.toml / setup.py — look for a [project.scripts] entry or a common entrypoint. If the project uses uv, prefer uv run <entrypoint>. If it uses plain Python, use python -m <module> or the script name.
3. Makefile — use make dev if the target exists, otherwise make run or make serve.
4. Procfile — parse the web: line and use that command.
5. docker-compose.yml / compose.yml — use docker compose up only if no lighter-weight option was found above.

If nothing is found, tell the user and ask them to specify the command. Do not guess.

Once you have the command, tell the user: "Detected start command: <command>" before proceeding.

---

Step 2 — Kill any existing instance

Kill any running process that matches the detected command so the new instance starts clean. Use pkill -f with a distinctive substring from the command (e.g., pkill -f navidrome-feed, pkill -f 'npm run dev'). Follow with sleep 1 to let the port release. Suppress errors from pkill with 2>/dev/null.

---

Step 3 — Start the server in the background

Run the server with output redirected to /tmp/dev-server.log (truncate first so old logs don't accumulate):

truncate -s 0 /tmp/dev-server.log
<start-command> >> /tmp/dev-server.log 2>&1

Use run_in_background: true so the Bash tool doesn't block.

Wait 4–5 seconds, then tail the last 15 lines of /tmp/dev-server.log and confirm the server started cleanly (look for a "started", "listening", "running", or "Application started" line). If you see an immediate crash or import error, report it and fix it before continuing.

---

Step 4 — Monitor only if issues were found

Do not start a Monitor unconditionally. Instead:

1. Check the startup tail from Step 3 for any ERROR, WARNING, Traceback, or Exception lines.
2. If the startup log is clean — tell the user "Server is running cleanly. No monitor started — I'll only look at logs if you report a problem." Stop here.
3. If the startup log contains errors or warnings — start exactly one Monitor (persistent):

tail -f /tmp/dev-server.log | grep --line-buffered -E "ERROR|WARNING|Error|Traceback|Exception|CRITICAL|failed|❌"

Set persistent: true. Tell the user what error was found and that the monitor is now active.

---

Notes

• Log file is always /tmp/dev-server.log so it's easy to tail manually too.
• If the server needs env vars (.env file), make sure the shell inherits them — uv run and npm run pick up .env automatically; for raw Python you may need dotenv run or to source .env first.
• Only start one Monitor per session. If one is already running (check if you've already spawned it this session), skip Step 4.