Skip to main content

Board and runs

The board is where Spark turns intent into durable work. A card can start as a rough request, become a routed task, attach to an agent run, and remain visible after the run has completed.

Cards
Hold launch intent, notes, labels, owners, and status.
Runs
Attach execution and streamed output to the card that caused it.
History
Keep routing decisions and lifecycle state queryable later.

Board model

Each project owns a kanban board with cards, labels, comments, assignees, and delivery state. Cards can be moved manually or by automation such as GitHub events, workflow completion, or close-out.

Squadboard aligns with upstream Squad's project-board model: the visible board is a projection over workflow state, and the Ready state is where work becomes eligible for automated pickup. The stored status is ready, so product language, APIs, and MCP tools use the same word.

Squad board stateSquadboard semanticWhat happens
BacklogbacklogCaptured or deferred work; not automatically dispatched.
Needs Researchcustom or a project-specific research columnClarification, spike, or Consult work before the card becomes Ready.
ReadyreadyCoordinator and Ralph-equivalent loops can pick up the card.
In Progressin_progressRuntime-owned; an active or pending run must exist.
Review / Donereview / doneHuman, peer, or ceremony review completes the inner loop before optional GitHub sync.

Starting work

For normal board work, put the card in Ready. The ready pickup loop starts the run automatically, moves the card to In Progress while the agent is active, then moves it to review when the run ends.

Manual starts are for explicit workflow/debug contexts outside the ready-column pickup loop:

  1. Open a card.
  2. Attach or run a workflow from the detail panel.
  3. Pick an agent only when you want to override routing.
  4. Watch the live run drawer.

Automatic pickup:

  1. Put a card in Ready.
  2. Ensure the project has active agents.
  3. Let the pickup sweep and coordinator create an issue_run.

Outputs

Outputs are the artifacts a run produced for the card: files, links, notes, or structured JSON. They are separate from the run log. Use the run log to understand what happened minute by minute; use Outputs when you want the result that should be reviewed, approved, linked from a ceremony, or carried forward into the next card.

Run states

StateMeaning
pendingRun is queued and waiting for execution
runningWorker is active and sending heartbeats
completedOutput was recorded and state advanced
failedRun failed or lease expired
cancelledRun was stopped intentionally

Workspace modes

Squadboard can run work in scratch space, a directory workspace, or a git worktree. Worktree-backed runs include branch and cleanup metadata when available.

Audit trail

Routing decisions, skipped dispatches, monitor decisions, close-outs, and state transitions should be visible as durable records rather than hidden prompt behavior.