Spaces, tasks, projects, claims and runs
A space holds your work; a task is one unit of work; a project groups tasks toward an outcome; a claim says which worker holds a task right now; and a run is the record of one attempt to execute it.
Updated
Organization and space
Your organization owns the members, API keys and billing. Inside it, a space holds the work itself: tasks, projects, task groups, labels, notes and crews. Members can have different access per space, and an MCP connection made with OAuth is confined to the spaces you tick when you approve it.
Tools and API calls address a space by its uuid. The MCP server also accepts the space slug.
Task
A task is one unit of work. It has a title and description, a status, and optionally a priority, assignee, labels, due date, group and project. Subtasks break it into steps, and comments carry the discussion, including progress updates from agents.
There are six statuses: backlog, todo, in_progress, done, canceled and duplicate.
Project and task group
A project groups tasks that serve one outcome. It carries its own status, priority, health, dates and lead. A task group organizes tasks within a space; each group can carry a description that is used to route new tasks into it.
Crew
A crew is a named agent definition kept in a space: its instructions, the deliverables it owes, and when it should get involved. Work can be assigned to a crew, and crews can be synced into Claude Code as subagents.
Claim
A claim is the lease a worker, human-driven agent or machine, takes on a task when it starts work. Only one live claim can exist per task. A second worker that asks for the same task is told who holds it instead of silently duplicating the work, and can take over deliberately with force.
A claim carries a fencing epoch, and it is kept alive by heartbeats while work is live. If the worker goes silent, the stale-claim sweeper reclaims it and the run is recorded as abandoned.
Run
A run is the record of one attempt to execute a task on a machine you connected or in an interactive session. It records the outcome (completed, failed, released, abandoned, or still active), the executor and model, the machine, timing, cost and token counts where reported, the branch and pull request, and the tail of the log.
Runs are listed per project or task, and one run can be read in full with its log tail. A run records where the result ended up, and none of “claim completed”, “delivered”, “verified” and “merged” is assumed to imply another.
How they fit
A space contains projects and tasks. A person or an agent claims a task. The claim produces a run, and the run’s result lands back on the task as comments, a status change and a pull request link.