The pipeline

The nine stages

The bar under the chat header shows Scope → Audit → Wireframe → Design → Architect → Build → Check → Test → Docs, each with its status: not started, working, waiting for your answers, needs you, needs attention (findings or failing tests), out of date (an earlier stage changed after it ran) or done.

Choose All 9 stages for the full board, with details, activity and checkpoints (it counts fewer if you’ve turned stages off). Select a stage to open its conversation.

Run pipeline

Run pipeline works through every stage that isn’t done. It creates Architect’s project files, then Build goes through Architect’s task list one task at a time.

Check’s findings go back to whoever owns the files, for up to two fix rounds. Failing tests get one fix round. Once Check passes on the code, the run does one visual check.

When it pauses

• An agent needs your answers. It carries on by itself once you’ve answered.
• Something goes wrong, like an error from the model.
• You switch to another project.
• The project has used its budget. Raise it in the cost meter to carry on.

Press Pause to stop the step that’s running straight away.

Automatic fixes

After Frontend, Backend or Test write files, Commandr runs your project’s typecheck script (or lint or build) and sends any errors back to the agent. Once typecheck passes, it runs your tests after Frontend and Backend and sends failures back — with rules not to delete, skip or weaken tests.

There are at most two fix rounds per run, and a round that changes nothing ends it. Turn either part off in Project settings → Repository → Automation.

After a restart

If Commandr restarts, or you switch projects mid-run, the run shows Review before resuming — it never restarts on its own. Look over the latest replies and checkpoints, then choose Resume pipeline.

After the build

Under each agent’s latest output, Run Check → Run Test → Run Document carry the work on. Debug and Sync are there when you need them — see Meet the agents.