The backlog.

One board holds the work: every idea, epic, and task, across every project you have open. Flows read from it and write back to it, so what Planner produces is what Sprint consumes.

01Ideas, epics, and tasks

Three entity types, linked by lineage. The type is fixed when the entity is created, so an item never changes from one to another.

Ideas are raw input. Yours, typed onto the board, or captured by a flow.
Epics are the grouping layer, added whenever an idea decomposes into more than one task. A large idea gets a full epic breakdown; any other multi-task idea gets one epic named after it. Only a single-task idea is epic-free.
Tasks are the executable unit. Every task knows its originating idea, and its parent epic when it has one, so you can trace a diff back to the idea that caused it.

Every entity carries the same two fields, plus one that only ideas have.

PriorityP0 through P6. New entities start at P2.

CategoryFeature, Bug, or Chore. New entities start at Feature.

Size hintIdeas only, and optional: small or large. It steers how a flow decomposes the idea.

An idea also takes file attachments. Paste or drop a file into the New idea dialog and it is copied to disk beside the idea rather than pasted into its text. Any file type, up to 10 MB each.

There are no side files and no parallel to-do lists. The board can mirror an external tracker, though: two-way sync ships for Linear, Plane, Dart, and Beads.

02Board stages

Five stages, four of them visible by default.

IdeaRaw input captured, not yet planned.

Ready for developmentApproved and queued. A flow drops approved tasks here; Sprint picks them up.

In developmentPulled into a live session. Derived by the orchestrator, not set by you.

DoneMerged and archived.

Won’t doDecided not to pursue. The column is hidden until you turn on the Archived toggle.

You cannot move a card into In development by hand. A task moves itself there while any run or sprint lane naming it is live, lands on Done only when that run’s work is merged, and otherwise returns to the stage it came from.

An epic’s stage is a rollup of its children. Every countable child at Done puts the epic at Done. One child short of that puts it back at Ready for development.

Children that are archived, parked at Won’t do, or still waiting on an approval gate are not counted. An epic with no countable children at all keeps the stage it has, and an epic you parked somewhere else by hand is left alone.

Ideas retire instead of moving.

When the Approve plan gate passes, an idea that produced epics or tasks is stamped retired and drops off the board, keeping the stage it was on. Its children carry the work from then on. An idea that produced nothing stays visible, which matters in a multi-idea batch where only some ideas get decomposed.

Note Epics and tasks a plan-gated flow drafts stay off the board and out of sprint batches until that gate approves them. Launch, Planner, and Ship carry that gate, and so does any custom flow with an Approve plan step. A flow without it writes to the board directly, and an idea is visible as soon as it is created.

03Working the board

Open it from Task backlog in the left rail, or press ⌘L, its default binding, which toggles the pane. The rail item carries the subtitle Planning pipeline and a count badge. Windows uses Ctrl in place of ⌘.

The board spans every project by default and narrows with the project dropdown. Kanban and List are the two layouts. Kanban is the default, and your choice survives a restart.

Search is a case-insensitive substring match over an item’s ref, title, and summary. It does not read the body, and it keeps an epic on screen when one of its children matches.
In sprint and In experiment narrow the board to specific batches or A/B experiments. A selection whose batch has ended still shows in the menu, marked stale, so you can clear it.
Archived reveals archived cards, dimmed and marked with an Archived chip, and adds the Won’t do column.
Sort offers Manual, Priority, Recently updated, and Title (A–Z). Manual is the default and none of the four is remembered between launches.

The New button opens a dialog headed New idea, and an idea is the only thing it creates. Epics and tasks come out of a flow’s decomposition.

Dragging re-ranks a card inside its own column. It works only while sort is Manual, with no search or membership filter on. Drag a card to a different column and nothing happens.

Stage moves live in the card’s ··· menu, under Change stage…, which lists only the stages you are allowed to set. That menu also holds Move up, Move down, Move to top, Show components, Archive or Unarchive, and Delete. Change stage, Archive, and Delete are disabled while the card has a live run, with the hint Finish or cancel the active run first.

A card leads with its project chip while the board spans every project. Then its type, priority, category, and size tags, flow and review markers, its display ref (IDEA-nnn, EPIC-nnn, TASK-nnn), title, summary, and age.

The footer carries a primary button and an Edit button, which opens that entity’s editor. On an epic or task that is title, summary, priority, category, and markdown body. On an idea it is title, summary, size hint, and markdown body. An epic also carries a task-count control that expands its children.

OpenOn an idea. Finds or creates that idea’s home session, a single persistent chat session with no worktree of its own.

RunOn an epic or task. Launches Planner for a planning-stage epic and Sprint for a task. On an epic at Ready for development or later it opens the batch picker instead, pre-ticked with that epic’s ready children, minus any already in flight and capped at the batch limit.

04The idea component ledger

Every idea tracks five components: Idea spec, Prototype, Architecture, Epics, and Stories. They render as chips on the idea card, in one of four states.

CompleteDone, with how long ago it landed.

Needs reviewPrior work exists, re-verify before trusting it.

Not startedNo work recorded.

SkippedDeliberately not done for this idea.

Needs review means the idea’s text changed underneath work that was already finished. The work is stale rather than wrong, so the flow should check it instead of redoing it.

Which components go stale depends on what you edited. An edit to the idea spec section re-flags prototype, architecture, epics, and stories. An edit confined to the architecture design section re-flags epics and stories only.

An edit the diff cannot attribute to either section re-flags all four, because the board flags rather than guesses. In every case, a component that was not Complete is left where it is.

Show components in the card menu expands a strip with one row per component, its provenance, and a manual override. You can set a row to Not started, Complete, or Skipped. You cannot set one to Needs review, and demoting a row out of Needs review asks you to confirm first, because that throws away the record that prior work exists.

Planner reads the ledger before it plans anything. If any component reads Complete, Skipped, or Needs review, it stops and asks you to pick: Resume, Redo a piece, or Start over. Only Resume skips finished work.

05Human tasks and dependencies

Some work has no agent that can do it. Opening an account, buying a device, signing something off.

A task carries an executor of Agent or Human, and flipping it to Human puts a neutral badge on the card. That control sits in the task detail overlay. The overlay opens from a task card in a run’s Decomposed stories artifact, or by following a card’s back-link to its originating idea and clicking the child.

A human task never becomes a sprint lane. It still lists in the batch picker, dimmed and chipped Human · runs outside the sprint, with its checkbox disabled. Nothing in a sprint moves it to Done.

Blocking prerequisites order a sprint: a task waits for the tasks it depends on.
Related edges are advisory. They record a connection and change no ordering.
A dependency on a human task is recorded but never blocks. The dependent still reads as ready, with neutral text such as waits on TASK-009 (human) rather than the orange blocked chip.

Start a sprint over tasks that depend on human work and it files one non-blocking item on the review queue for each human prerequisite, then proceeds. The item is keyed to the human task, so a second sprint depending on the same one files no second copy. A human task that is archived, or already at Done or Won’t do, is skipped.

Tip Agents create dependency edges through their own tooling. The one dependency control in the app is removing a prerequisite from the Depends on list in that same overlay. If the ordering looks wrong, that is where you fix it.

06The project overview

Click a project in the sidebar and you get its own home page. The default layout is three sections, in this order: Active agents, Recommended actions, and Backlog.

The Backlog section shows one tile per visible board stage, so four tiles, and two sub-lists underneath: Top ideas and Next up · Ready for development. Each sub-list has its own launcher bar, Launch planner and Launch sprint.

The Recommended actions card Launch a sprint appears whenever at least one task is Ready for development and free. Its Select tasks button opens the sprint batch picker. The picker groups tasks by parent epic in collapsible groups, each group with a tri-state checkbox that selects or clears the whole epic and a live count of what is picked.

Tasks already in flight cannot be selected and do not count toward a group. A flat backlog with no epics renders as a plain list.

View dismissed counts and lists the recommended actions you dismissed, each with its own Restore control. A card is listed only while it still applies, and dismissals are per project.

This page is one of the two surfaces you can customize, along with the review queue. A custom view reorders the three sections, hides any of them, and puts widgets between them. See customizing the queue.