Backlog Column Prompts

  • Desktop, mobile and web
  • All plans

What it does

Gives each backlog column its own job. You write a prompt for a column ("Analyse this ticket without coding it, then explain what is unclear", "Write the implementation plan", "Review the changes"...) and, optionally, the agent that runs it. When a ticket is dropped in that column, the prompt is sent with the ticket: either to the agent already working on it, or to a new agent. Works on every column, built-in (TODO, Development, Pending, In Review, Done) and custom.

Where to find it

  • Desktop, inside a project's Backlog: the Column prompt button in every column header (next to ...). It is highlighted when the column has a prompt; click it to add or edit one. The same window is also in ... (Column options) > Column prompt.
  • Until you save your first column prompt, empty columns show an arrow up to that button: Customize the prompt, sent when a ticket lands in this column. It goes away for good once a prompt is saved, and never shows on a board that already has one.
  • When creating a column: New column > Advanced options sets the same prompt, mode and agent, so the column is born with its job.

How to use it

  1. Open Column prompt on a column, write the Prompt. On Development, the field starts filled with the default instruction ("Implement this ticket end to end..."): without a saved prompt, the agent receives the ticket and implements it, and this text does the same, so edit it to change what Development asks.
  2. Choose When a ticket is dropped here:
    • Ask me (default): each time, a small window offers Keep the running agent, New agent or Just move the ticket.
    • Keep the running agent: the agent already working on the ticket receives the prompt. If none is running, a new one starts.
    • Start a new agent: a fresh agent runs the prompt. The previous one finishes its turn, then closes.
    • Do nothing: the ticket only changes column. No agent starts and none receives the prompt; the text stays saved for later. Not offered on Development, which always starts the ticket's agent. This is how Pending comes in the board templates.
  3. Under Agent for this column, choose The ticket's agent (default: the agent working on the ticket, otherwise the one assigned to it, runs the prompt), The ticket's initial agent (always the agent assigned to the ticket: if a previous column put another agent on it, that agent is replaced instead of receiving the prompt), Reclassify (adaptive model) (the AI reads the column's prompt and the ticket, picks the agent and the model, and saves them on the ticket before the launch; one AI call per launch, models excluded in Settings > Agents & AI > Optimization are never picked; team tickets are left as they are) or A specific agent, then pick it in the agent picker (avatars, search: the project's agents first, then roles). With a specific agent, Use this agent even if the ticket has one (on by default) decides who wins when the ticket already has an assigned agent; off, the column's agent is only used for tickets without one. Under the picked agent, the Provider, Model and Reasoning cards (plus Account for providers with several accounts) set what that agent runs on, like in Add Agent; each starts on Inherit (the agent's own config). They apply only when the column's agent is the one launched, and pin over the ticket's own provider / model.
  4. Save. Remove the prompt turns the column back into a plain column.

Tickets moved by agents (MCP) also run the column prompt, without opening any window: an Ask me column then keeps the running agent, or starts a new one when none runs. Agents started this way open in the background, they never steal focus.

Settings

  • backlogColumnAutoLaunchLimit (global, Settings > Backlog > Automatic column prompts per ticket): how many column prompts moves made by agents can start on one ticket in 24 hours, so two agents cannot bounce a ticket forever. Default 10, 0 = unlimited. Moves you make yourself are never limited.
  • backlogColumnAutoLaunchLimit (project override, the project's Backlog settings): a different limit for one project; Use the global value goes back to the global one.

The prompts themselves are stored with the board on the server, so every computer and the phone share them.

Agent tools (MCP)

  • backlog_update / backlog_start_next: moving a ticket into a column with a prompt runs that prompt (subject to the loop limit).
  • backlog_update / backlog_reorder moving a ticket into In Progress starts its agent like a drag on the board (its worktree included), even when In Progress has no prompt. Nothing new starts when a console already works the ticket or when the agent moving it is the one assigned to it. The answer says what started (agentStart).
  • settings_get / settings_set with backlogColumnAutoLaunchLimit (global or project scope).
  • backlog_columns_list and backlog_column_save: read and change the prompt, mode, agent, strategy, priority over the ticket's agent, provider, model, effort and account of any column, built-in or custom. Only the fields passed change; prompt: "" alone removes the prompt. See Custom Backlog Columns.

Providers

All providers. The prompt is sent as a message to the running agent, or as the first message of a new one, whatever its CLI.

Mobile

When you move a ticket on the phone into a column with a prompt, the same choice (Keep the running agent, New agent, Just move the ticket) appears in a bottom sheet for Ask me columns; the desktop runs it. Writing or editing a column prompt is done on the desktop, like creating columns. Needs the desktop to be connected.

Limits

  • One prompt per column. Not on the idea box.
  • Closing a ticket from an agent's own Done button, the cleanup window or a team run does not run the Done column's prompt; dragging it to Done on the board does.
  • The loop limit only counts moves made by agents.

Common questions

  • Can a backlog column start an agent with its own instructions? Yes: the Column prompt button of the column header, write what the agent should do with tickets dropped there.
  • Moving a ticket to Pending starts an agent, how do I stop that? Open the Column prompt of Pending and choose Do nothing (or Remove the prompt). The board templates now set Pending to Do nothing; switch it to Keep the running agent if you want the agent at work to sum up where it stands when you park its ticket.
  • Can I keep the same agent when I move a ticket to the next column? Yes: choose Keep the running agent (in the column, or each time with Ask me). It receives the new column's prompt and keeps its context.
  • What happens to the old agent when a new one starts? It finishes its current turn, then closes.
  • What happens to the previous column's agent when the ticket moves to a column run by a team? Same thing: when the team run starts, the agent leaves the ticket, finishes its current turn, then closes. An agent waiting for your answer is not closed.
  • What does Development send when it has no prompt? The ticket alone, which the agent implements. The Column prompt window of Development shows that default as editable text; saving it keeps the same behavior until you change the words.
  • Can I set the prompt while creating the column? Yes: New column > Advanced options.
  • Can I force a specific agent for a column? Yes: Agent for this column > A specific agent, pick it, and keep Use this agent even if the ticket has one on to override the ticket's own agent.
  • Can a column go back to the ticket's own agent after another column used a different one? Yes: Agent for this column > The ticket's initial agent. If another agent is working on the ticket, a new one with the ticket's agent replaces it.
  • Can the AI choose the agent and the model for a column? Yes: Agent for this column > Reclassify (adaptive model). The choice is saved on the ticket, so the next columns and the card follow it.
  • How do I see that another agent than the ticket's is working on it? The card shows it next to the ticket's agent, with its model; hover it to see which column step it runs. Same on the phone.
  • Can a column run its agent on another model or account? Yes: with A specific agent, use the Provider, Model, Reasoning and Account cards under it. Inherit keeps the agent's own config.
  • Does a column's model replace the model pinned on my ticket? Only when the column's agent is the one that runs: with Use this agent even if the ticket has one on, or on a ticket without an agent. With it off, a ticket that has an agent keeps that agent and its own provider, model and reasoning, even when it is the same agent as the column's.
  • My ticket ran on another model than the one it pins, how do I find out why? The agent's console shows a line when it starts, and the ticket gets a comment labelled Launch check that names the cause: the column's engine, a temporary launch override (Usage panel), a quota rule, the CLI options of the agent or provider, or a model this computer cannot use. Nothing is changed for you: each of these is a setting you can edit. A later launch back on the ticket's model says so in a new comment.
  • Can an agent check this over MCP? backlog_get: effectiveExecutor takes the columns into account (modelSource: "lane" when a column imposes its engine), and validation repeats the last Launch check. A temporary override, a quota rule or CLI options depend on the computer that launches, so they only show once the launch happened.
  • My agents keep moving a ticket back and forth, how do I stop it? Agent moves start at most 10 column prompts per ticket per 24 hours by default; lower it in Settings > Backlog > Automatic column prompts per ticket.