# Agents working on a board

Give agents instructions, see what they are working on, and follow along.

An agent such as Claude Code can keep a project board up to date while it works: it reads the board, picks an item, and moves it along as the work gets done.

## What agents do by default

Every agent that connects to Listspace is told to keep your board up to date without being asked, and most agents do: they read the info lists and the summaries earlier conversations left first, add tasks, ideas, questions and, where decisions are on, decisions as items when they come up (unless one already covers them), put a new task in your to-do list and in a doing list only once the work has started, keep notes, documentation and decisions in an info list, claim an item and move it to the list that fits when work on it starts, move it again with a short comment on what they did when it is finished, release an item they stop without finishing, update the matching items when you tell them about a call, an email or a meeting, ask you one short question about items that look finished or have gone quiet, keep items few and clear, update the board as things happen, and leave a short summary when they finish. When the work is also tracked somewhere else, such as a pull request or a deploy, they still move the matching item and comment on it. A board made from a template also comes with instructions for its kind of work, such as groceries, a client project or meeting notes. The board's agent access still decides what happens directly and what waits in the Inbox.

## Keep a project's board up to date

For a coding project, you can add one line to the project's CLAUDE.md (Claude Code) or AGENTS.md (Cursor, Codex and most others). Every agent that works in that folder reads it:

`Keep this project's Listspace board up to date as you work: move items when you start and finish, comment on what you did, and record decisions.`

The Listspace plugin for Claude Code also checks at the end of each turn: when Claude changed files, committed, opened a pull request or deployed and told the board nothing, it asks Claude once to move the item and add a short comment.

## Tell agents how to work

- **Guidance for agents**, in **Settings**, then **Your AI**, is read first on every board. Use it for how you like to work in general. Until you write your own, agents follow the default shown there; your own guidance replaces it.
- **Instructions for agents**, in the board's **⋯** menu under **Board settings** > **Agent access**, are for one board. For example: "Put bugs in Bugs, and keep titles short".
- **Fill instructions for agents**, in a field's **...** menu in the board's **Fields**, tell your AI how to fill that one field, for example "Set from the latest call notes". Only the board owner or a team admin sets them. See [Fill instructions for agents](https://listspace.app/help/custom-fields#fill-instructions-for-agents).

An agent gets both when it starts a session on a board, together with its agent access level, the lists, the open work and the summaries the last three sessions left there. At the end of a session it leaves a short summary on the board (what changed, what is open, the next step), so the next session, by the same agent or another one, picks up from it.

## Claimed items

An agent can claim an item so other agents leave it alone. The item then shows "Claude is working on this". In the item window you see until when, and you can **Release** the claim at any time. A claim ends by itself after a while, two hours unless the agent asks for another time.

An agent that puts an item in your Doing list, by creating it there or moving it there, claims it at the same time, so you always see who is working on what. When another agent is already working on it, that claim stays. Moving the item out of Doing ends the agent's claim.

Claiming also moves the item to your Doing list, so an item an agent is working on never sits in To do. That happens when the item is in a to-do, backlog or capture list and the board has a Doing list (with several, the first one), and only when the board's agent access lets the agent move items. Releasing the claim leaves the item where it is: when the work is finished, the agent moves it to your Ready for review list when the board has one, or to Done, with a comment.

Finishing an item ends its claim. When the item moves to a Done list, its due date is marked done or it is archived, "Claude is working on this" goes away, whoever made the change: the agent, another agent, or you in the app. Undo moves the item back but does not bring the claim back.

An agent can also ask for the next item: it gets the top open item from the to-do lists, then the backlog, that no other agent holds. Claiming and taking the next item need agent access level 4 (Act) or higher.

## When an agent forgets to update the board

In a long session an agent can lose track of the board, most of all after its conversation is summarized to save space. Listspace keeps it on track, in every app that connects to it:

- **Reminders while it works.** When an agent has claimed an item, or put one in Doing, and has not moved it, changed it or commented on it for half an hour, its next answer from Listspace reminds it: move the item to Done with a comment when the work is finished, or comment on where it stands. The reminder names the items and repeats the board rule in one sentence. It comes at most once every half hour per item, only to the agent that did the work, and never to a read-only connection.
- **The next session checks first.** When an agent starts a session on the board, it gets the items earlier sessions of your AI left in Doing (or claimed) and quiet for half an hour or more. It checks whether that work was done, for example in the code history, moves what is finished to Done with a comment, and asks you only when it cannot tell. So the board is put right before you see an item that looks stuck.
- **The end of a session.** When an agent leaves its session summary, Listspace lists the items it still has open on the board, so it can finish them or say where they stand.
- **Items that wait on someone.** Some work cannot move until someone acts, for example your review or your merge. The agent then says what the item waits for, and the item shows "Waiting for: your review" in a muted line. The reminders leave that item alone until anything changes on it: you move it, edit it, comment, tick a checklist item or mark it done. After 7 days without a change the next session asks itself whether it still waits.

## No doubles when a connection drops

Sometimes an agent loses the answer to a change: the connection drops, or Listspace restarts for an update and answers with an error, although the change was made. A plain retry would then add the same item twice. Listspace prevents that in two ways:

- **A key per change.** Agents can send a key of their own with every call that adds something (an item, a comment, a list, a checklist, a file, a decision). When they retry with the same key within 24 hours, they get the first answer back and nothing is added again.
- **A one-minute check.** Without a key, the same item (same list, title and description) or the same comment on the same item from the same agent within a minute is taken as a retry: the agent gets the first one back instead of a copy.

## Follow along

- The **Inbox** column on the board holds what agents suggest. See [The Inbox and agent access](https://listspace.app/help/inbox-and-agent-access).
- **History**, in the board's **⋯** menu, shows every change, with who made it, and starts with a summary since your last visit and an **Agents** section: each agent that works on the board, with the items it is working on, what it waits on you for (a request or a suggestion in the Inbox) and what it finished in the last day. On Free it shows the last 30 days. Show only one person or agent, and undo an agent's change. When you open a board after 3 days or more away, a short note at the top, **Since your last visit**, sums it up and links to History; close it with the cross. Ask your AI to "catch me up" and it gets the same summary.
- **Agent access**, in the board's **⋯** menu under **Board settings**, shows the board's level, for example **Capture (3)** or **Follows account (Capture)**. The menu shows it too, under **Board settings**. Choose **Change** to change the level or the board's instructions.
- **Webhooks**, in **Settings** then **Capture and connections**, tell your own tools when items are added, moved, changed or completed, when a suggestion comes in or is decided, and when a comment is added. Webhooks are part of Pro. Agents cannot see or add webhooks.

Everything an agent may do is listed in [What agents can do](https://listspace.app/docs/agents).

## See the guides

- [Choose how much your AI may change](https://listspace.app/guides/choose-how-much-your-ai-may-change)
- [Plan in ChatGPT, build with Claude Code: one board for both](https://listspace.app/guides/plan-in-chatgpt-build-with-claude-code)
- [Use Listspace with Claude Projects](https://listspace.app/guides/use-listspace-with-claude-projects)
- [Board instructions and fields: what goes where](https://listspace.app/guides/board-instructions-and-fields)

This page as HTML: https://listspace.app/help/agents-on-a-board. As markdown: https://listspace.app/help/agents-on-a-board.md.
