Blog Dashboard

Blog Dashboard

Start or stop the local Flask dashboard that visualizes the blog pipeline Kanban board. The dashboard itself never runs skills directly — it just watches blog-output/ and dashboard/state.json and spawns agents via queue.json. See dashboard/README.md for the full architecture.

Resolve <plugin-root> the same way other dashboard-touching skills do: the directory containing dashboard/app.py, either the current plugin root or CLAUDE_PLUGIN_ROOT if set.

The action is the first argument (start or stop). If missing or not one of those two, ask the user which one they meant — do not guess.

start

1. Check if it's already running:

   lsof -i :5001 -sTCP:LISTEN -t
   

   If this returns a PID, tell the user the dashboard is already running at http://localhost:5001 and stop — do not start a second instance.

2. Ensure the virtualenv exists. If <plugin-root>/dashboard/.venv is missing, create it and install deps:

   cd <plugin-root>/dashboard && python3 -m venv .venv && .venv/bin/pip install -r requirements.txt
   

3. Launch the app in the background, logging output and capturing its PID so stop can find it later. Put the cd on its own line — chaining it as cd dir && nohup ... & backgrounds the whole cd && nohup pair in a subshell, so the following echo $! line runs in the wrong directory and writes the pidfile to the wrong place:

   cd <plugin-root>/dashboard
   nohup .venv/bin/python app.py > .dashboard.log 2>&1 &
   echo $! > .dashboard.pid
   

   Run this with the Bash tool's run_in_background unset (it's already detached via nohup ... &) — just run it and move on, don't block on it.

4. Give the server a moment to bind, then confirm:

   sleep 1 && lsof -i :5001 -sTCP:LISTEN -t
   

   If a PID comes back, tell the user the dashboard is running at http://localhost:5001. If not, read the last ~20 lines of dashboard/.dashboard.log and show the user what went wrong (commonly: missing venv deps, or missing WP_APP_PASSWORD/GEMINI_API_KEY env vars — see dashboard/README.md prerequisites).

stop

1. Find the listening process:

   lsof -i :5001 -sTCP:LISTEN -t
   

   If nothing is returned, tell the user the dashboard isn't running and stop.

2. Kill it:

   kill $(lsof -i :5001 -sTCP:LISTEN -t)
   

3. Clean up the pidfile if present:

   rm -f <plugin-root>/dashboard/.dashboard.pid
   

4. Confirm to the user that the dashboard has been stopped.

Notes

• Always resolve the port (5001) by checking lsof, not just the pidfile — the pidfile can go stale if the process was killed out-of-band or the machine restarted.
• Never touch dashboard/state.json, queue.json, or blog-output/ from this skill — that's /blog-dashboard-cleanup's job, and normal pipeline operation otherwise.