Your First Mission Control: Automated Task Board + Projects

Open your terminal and run this single command. It clones the repo, installs everything, and starts the server in one go:

git clone https://github.com/josephtandle/mastermind-mission-control && cd mastermind-mission-control && bash install.sh

The install script handles everything: dependencies, database setup, and starting the server. It takes about two minutes the first time.

After the install, the task executor needs to know where Claude lives on your machine. If it was not picked up automatically, paste this into Claude Code to fix it:

Find the Claude binary and wire up the task executor to use it.

1. Locate the claude executable on this machine. Try these in order:
   - Run which claude (Mac/Linux) or where claude (Windows)
   - Check common install paths: ~/.local/bin/claude, /usr/local/bin/claude, %APPDATA%\npm\claude.cmd (Windows)
   - If none found, check the output of npm list -g claude or npm root -g
2. Open executor.py in this project directory.
3. Find the subprocess.run call that invokes claude. Replace the bare "claude" string with the full absolute path you found in step 1.
4. Save the file.
5. Test it by running python3 executor.py (or python executor.py on Windows). If a card is in the AI Auto Execute column, it should run. If the column is empty, confirm the script exits cleanly with "No pending cards in AI column."
6. Write DONE when the executor runs without a "not found" error.

Deploy when you're done.

Once the install finishes, go to this address in your browser:

Any time you want Mission Control open again, just go back to that address. Your board and all your cards will still be there. If the server is not running, open your terminal, go to the mastermind-mission-control folder, and run npm start.

Paste this into Claude Code to verify the file browser paths and lazy loading are working correctly on your machine:

Wire up the Mission Control file browser — verify paths and lazy loading.

You are setting up the Mission Control file browser on this machine. Complete all steps.

STEP 1 — DETECT HOME DIRECTORY
Run this and note the output:
  Mac/Linux: python3 -c "import os; print(os.path.expanduser('~'))"
  Windows:   python -c "import os; print(os.path.expanduser('~'))"

STEP 2 — VERIFY INITIAL LOAD
Run:
  curl -s -o /dev/null -w "%{http_code} | %{size_download} bytes | %{time_total}s" "http://localhost:3001/api/repo/tree?depth=2"

Expected: 200 | under 200000 bytes | under 2 seconds
If the server is not running, start it first: cd mission-control && npm run dev

STEP 3 — VERIFY LAZY EXPAND WORKS
Pick any subdirectory from the home directory and test the expand endpoint:
  Mac/Linux: DIR=$(python3 -c "import os; print(os.path.expanduser('~/Desktop'))"); curl -s -o /dev/null -w "%{http_code} | %{size_download} bytes" "http://localhost:3001/api/repo/tree?path=$(python3 -c 'import urllib.parse, os; print(urllib.parse.quote(os.path.expanduser("~/Desktop")))')&depth=1"
  Windows: Use Postman or browser: http://localhost:3001/api/repo/tree?path=C%3A%5CUsers%5CYOURNAME%5CDesktop&depth=1

Expected: 200 | under 50000 bytes

STEP 4 — FIX EXECUTOR PATH
Find the claude binary and update executor.py:
  Mac/Linux: which claude
  Windows:   where claude

Open mission-control/executor.py. Find the subprocess.run call. Replace "claude" with the full path you just found. Save the file.

STEP 5 — CONFIRM EVERYTHING
Run the executor with no pending cards to confirm it starts cleanly:
  python3 executor.py   (Mac/Linux)
  python executor.py    (Windows)

Expected last line: "No pending cards in AI column"

Write DONE when all 5 steps pass. If any step fails, write the error and NEEDS_REVIEW.

---
How the two-level loading works:
- Depth 2 on initial load — you see your top-level folders and their immediate contents without clicking anything. No blank tree, no spinner on first open.
- Depth 1 on expand — when you click into a folder that has not been loaded yet, it fetches just that folder's direct children. One network request per expand, each one tiny.
- Already-loaded nodes — if children are already in the tree from the depth-2 initial load, expanding is instant with no network call at all.

The five pre-loaded sample projects are:

Each project has cards distributed across the columns already. Click through them and read what is in each card. You will start to see how a real workflow is supposed to look.

At the top of the board there is a projects dropdown. Click it and select any project to filter the board to show only that project’s cards. Click it again and select “All Projects” to go back to seeing everything.

To create a new project: click the Projects dropdown at the top of the board, then select + New Project. Give it a name that represents something real in your business.

Once you have explored the sample cards and understand how the board works, you can remove them all at once: go to Settings and choose Clear Sample Cards.

Do not do this yet if you are still in the session. Wait until you have added at least one real card first so the board is not empty.

Click the + button at the top of the Backlog column. A new card form will open. Give it a title that describes the task clearly enough that you will know exactly what it means three weeks from now.

Inside the card, choose a project from the dropdown. Then add a description: a sentence or two about what done looks like, any context Claude will need later, and any relevant links or notes.

You can also add labels to tag the card. Labels are optional but useful for filtering later: things like content, client, or marketing.

Click and hold the card, then drag it to a different column. Try moving it to Doing, then to Review, then back to Backlog. That is how work moves through the board.

Cards can also be moved by opening them and changing the status field in the card detail view. Drag and drop is faster for quick moves. The status dropdown is better when you are already reading the card.

The sample data includes cards already in the AI (Auto Execute) column. Open a few of them and read the titles and descriptions. These are real examples of tasks that Claude can execute fully automatically once the integration is wired up.

As you read them, think about your own business: what tasks do you do every week that follow a predictable pattern? Those are your best candidates for this column.

You are going to give Claude a one-time instruction so that whenever you say “add this to the MC task board”, it knows exactly what to do: create a card in Mission Control with the right title, description, and column, without you having to explain it every time.

Paste this into Claude Code. It adds a permanent note to your CLAUDE.md file, which is the file Claude reads at the start of every session to understand your setup:

Add the following to my CLAUDE.md file:

## Mission Control Task Board

My personal task board runs at http://localhost:3001.

When I say "add this to the MC task board" or "add a card to Mission Control":
- Create a new card via POST http://localhost:3001/api/cards
- Body: { "title": "[card title]", "description": "[description]", "status": "backlog" }
- Put it in Backlog unless I specify a different column
- Confirm when done and tell me the card title

After Claude adds this, you never have to explain the board again. It becomes part of your permanent setup.

The board already has sample AI tasks to show you the format. Now add one of your own. Paste this into Claude Code, fill in your details, and it will create the card:

Once the card is on the board, the task executor picks it up automatically and runs it with Claude Code. It checks the AI column once an hour. When it finishes, your card moves to Done or Review on its own.

You can also chain research into action. This prompt researches, installs, and logs the result to your board in one shot: