ADHD, six terminals, and a list you don't have to keep

A friend of mine has ADHD and builds software with coding agents. The last time we talked about work, they said the agents had made their days harder to follow, not easier. This is their complaint as well as I can retell it, and then what I would put in front of them.

The complaint

They sit down to fix one thing and start a session for it. Halfway through they notice something else, and because starting an agent costs nothing, they open a second terminal and start that too. Then a third idea arrives, and it would be a waste not to.

By lunch there are six terminals. One finished an hour ago and nobody looked. One has been waiting for a yes or a no since ten. Two are probably on the same thing, because they asked for it twice in different words. The thing they sat down for in the morning isn't done.

Before agents, their head could only start as much as their hands could type. Now starting is free, and keeping track is the only job left. It is the one job they can't do.

They have tried the systems: a board, a TODO.md, a notebook beside the keyboard. Each worked for about four days. All of them ask for the same thing, which is to stop what you are doing and write it down. For my friend, that is the first thing to go the moment the work gets interesting.

A list doesn't change how anyone's head works, and this post isn't advice about ADHD. But the complaint has a shape I know, because it is the one Hamloy is built around: the list shouldn't depend on the person who is deep in something else. The ones doing the work should keep it, and you should be able to read all of it on one page.

One page instead of six terminals

Every agent connected to a project writes to the same list, and the project's board shows it:

Here now, beside every list, is each session with its repo, its branch, the task it is on and when it last did something. Recent changes is what happened, newest first, with who did it. That is the question my friend can't answer at lunch, answered without opening a single terminal.

The overview has a page for each of the other questions:

• Right now: who is working, on what, and what is ready with nobody on it.
• For you: what only you can do or decide, the one that holds up the most work first.
• Recently done: what was finished, day by day.
• Gone quiet: open items nothing has happened to for a week. The thing you started and forgot is here, by name.
• History: every change, who made it and when.

The agents keep it, not you

Nothing on that page was typed by you. Each agent is told, by the server's own instructions, to claim a task before it starts and tick it when it is done. You don't write that rule into a file, and you don't update anything afterwards.

⏺ hamloy - Claim an item (MCP)(key: "B-k3f9")
  ⎿ B-k3f9@4 "Streak counter on the home screen" claimed by this session

So when you type "fix the streak counter" into terminal four, the list knows before you have switched windows. And if terminal two was already on it, because you asked this morning in other words, terminal four is told so and offered something else:

• Called hamloy.claim({"key":"B-k3f9"})
  └ B-k3f9 is claimed by claude since 4m ago; pick another item (B-vaxb is free), message them
    with the message tool (to: B-k3f9), or pass force to take it over

The idea that arrives halfway is one sentence to the agent you are already talking to: "put a home screen widget on the list for later."

⏺ hamloy - Add an item (MCP)(title: "Widget for the home screen", section: "Later")
  ⎿  B-paz4@2   Widget for the home screen

The idea is kept, and you didn't open a terminal for it. What sits under a heading named Later stays out of the agents' way too: they hear only how many items it holds.

You walked away, and the work is still there

A session that closes lets go of its task at once, and one that has been quiet for half an hour does too, so a task never stays held by a terminal you forgot. When you come back and start a new session in the same folder, it sees what the earlier one was doing:

[ ] B-k3f9@4   Streak counter on the home screen
      (claimed by your earlier session 2d75bc67, quiet since this one began)

It takes the task back without asking you anything:

B-k3f9@4 claimed by this session, taken over from your earlier session 2d75bc67,
  quiet since this one began

The notes the earlier session left under the task come with the claim: what it tried, and what it ruled out. You don't have to remember where it was. Handing over a task with no handoff file shows two of these word for word.

Look at it the way your head wants today

A backlog in the order it was written is a wall of text, and my friend's head wants something different from it every day. Some mornings it is "only the important ones". Some afternoons it is "what can I start right now". A view is a saved way to see a list, as rows, as a board with columns, or as a table, and it is a tab on the board.

You can make one by asking. "Show me the tasks as a board by priority":

view(name: "By priority", layout: "board", group: "prio", order: ["high", "medium", "low"])

V-yqtx@1  By priority: board of backlog, grouped by prio (high, medium, low)
  shows: Board of backlog: 8 of 8 tasks; columns by prio: high 1, medium 1, low 1, no prio 5

"Just the most urgent, as a table":

view(name: "Most urgent", layout: "table", sort: ["-prio"], show: ["prio", "area"])

V-ed8w@1  Most urgent: table of backlog, sorted by -prio, showing prio, area
  shows: Table of backlog: 8 of 8 tasks; sorted by prio high to low (most important or newest first); showing prio, area

Or press + New view on the board and pick a ready-made one: "Ready to start", "Who is on what", "Waiting on something", "Recently done". Each becomes a tab, and going from one way of looking to another is one click. Nothing is moved or sorted by hand, because a view only changes what you see. The list under it stays as the agents keep it.

An agent can also suggest a view. It shows as a card above the tabs, "Codex suggests a view: Waiting", with what it shows and why, and three buttons: Look, Dismiss and Keep it.

A decision is one click

The terminal that has been waiting since ten was waiting for an answer. With a shared list, the agent doesn't stop and wait in its own window. It adds a question, with the options and the one it recommends, and goes on with what doesn't depend on it. The question is on For you:

You pick one, and the tasks that waited on it are ready. In Claude Code, the status line says how many things wait on you, so you see it in the window you are already in. Across several projects you can ask instead: "What's waiting on me?" gets one answer for all of them.

What did I do today?

It is the question at the end of a day spent in six places at once. Recently done answers it day by day, and History by the minute, with who did each thing. Read your assistant's night shift shows that page after a night of work.

What you keep doing

You still start the agents, as many as you like and whenever the idea comes. You still decide. What you no longer do is the bookkeeping: which one is on what, which is done, which is waiting on you, and where you left off.

Try it

One project is free, with as many agents as you like on it. Connecting takes one line in the project's folder; the steps are in the docs for Claude Code and Codex. The pages this post leans on are the overview, Views, Claims and Questions.