Skip to content

Work a project's task list

A project’s Tasks tab is every task that belongs to the project, in one place. It is built for two jobs: seeing where the work stands, and moving a lot of it at once.

The row of chips at the top is the map. One chip per status the project actually has, each with its count:

Blocked 2 · In progress 3 · Open 14 · Backlog 8 · Done 41

Those counts cover the whole project, not what is on screen, so they stay a useful answer to “what is in here” no matter how far you have filtered. Selecting a chip filters the list to that status; selecting it again clears it.

Below that, the list is grouped by status by default, blocked first, because blocked is the only status that is asking you for something. Each section states its status once and can be collapsed. Turn Group by status off for one flat list sorted by most recently touched, and the status moves into its own column.

Finished work is hidden until you ask for it. Show closed brings done and cancelled back in.

An agent that finds real work outside its scope proposes it here instead of doing it. A proposal starts nothing on its own: like every other backlog item, it waits for a person to move it out.

ControlNarrows by
SearchTitle, task key, or intent
Any assigneeThe agent, person, or position holding the task, or Unassigned
Any repositoryThe repository the task is bound to, including Not set and No repository
Any typeThe task type, when the project has more than one

Every filter is written into the page address, so a filtered view survives a reload, comes back when you press Back, and can be sent to someone else as a link.

The line under the filters says how much you are looking at: “Showing 11 of 48 tasks”, with Clear filters beside it whenever anything is narrowing the list.

  1. Select rows with their checkboxes. A section heading’s checkbox takes the whole status at once, which is how you assign everything blocked or send a whole backlog back.

  2. The action bar appears at the top of the list and stays there while you scroll. It names the size of what you selected: “3 tasks selected”.

  3. Choose the action.

ActionDoes
Assign N tasks to…Hands the selection to an agent or to a person. Clear the person assigned removes a personal assignee
Auto-assignScores the selection and gives each task to whichever agent the router picks
Set repository…Points the selection at one of the project’s connected repositories, or records that it needs none
Send to backlogParks the selection. Move to open brings parked work back

The selection only ever covers what is on screen. Filter something out and it drops out of the selection with it, so a bulk action can never reach a task you cannot see.

Pointing a task at a repository is not only bookkeeping. A task that was blocked purely because nobody had said which repository it belonged in is released by that answer - and if an agent is already assigned to it, the run starts.

The same is true of assigning an agent to a task that is ready to be picked up. When an action will start work, you are told before it happens:

Assign 3 tasks to Codey? 3 tasks will change, and 2 will start an agent run straight away.

Nothing that only re-labels a task asks first, so the question always means something.

A bulk action reports on every task you selected, and anything it did not change is listed under the filters with the reason, in words:

2 tasks were not changed SUM-14 SKIPPED A setup task is a person’s to do, so it takes a person, not an agent. SUM-31 SKIPPED A content task has its own lifecycle. Move it from the task page.

The two common reasons:

  • Some work is not an agent’s. A setup or approval task exists to collect a decision or a value from a person, so an agent cannot be assigned to one.
  • A typed task owns its own lifecycle. A content task moves through states of its own, and backlog is not one of them. Move it from its own page instead.

The report stays until you dismiss it, so nothing you need to act on disappears on a timer.

Rows that have more to say carry a + on the right. Opening it shows the task’s intent, why the router picked the agent it did and how confident it was, and the other agents it considered. Selecting one of those alternatives assigns it.