Open full size ↗
Open full size ↗
Start the build
Approve and build here continues the App's planning session after its turn. A waiting terminal planner gets Build in <agent> · terminal and writes the Outline next. Otherwise Start the build opens on Plan.
Open full size ↗
Open full size ↗
Example task in a fictional project.
| On the card | What it picks |
|---|---|
| Agent | The coding agent for this build |
| Build model and Effort | The model and effort for the session that writes and tests the code, with the recommended model and why |
| Subagent model | Claude Code only, folded: the model for slices it hands to subagents; other agents choose their own |
| What's different | What changes for the chosen agent, for example questions as numbered choices |
| Files for the build and Workspace | Attachments, run location and branch/worktree choices |
Press Approve & start for an App run. In my terminal shows one request box with Copy and a folder chip; hover for the full path or click to copy it. Paste into any agent, or ask in your own words.
For a pending Plan, Approve and copy approves it and copies the request. After approval, Ready to build offers Copy prompt. Closing the sidebar keeps the run (agent runs).
Where the build lands
A new branch or worktree starts at this checkout's HEAD. A worktree's name sets its branch and folder beside the project. The terminal equivalent is qualitylayer branch new [name]. Pull requests need a branch other than their target. Pushes and deployments need the Plan's authorization; you control merging.
While the agent builds
Open full size ↗
Open full size ↗
Example task in a fictional project.
Implement shows the slices with their tasks: every slice open for a small build, the one in work for a larger one, and past 10 slices the finished ones in one row. A task opens to one sentence and its card (files, test first, done when). The header counts tasks and slices and shows a red chip only while a check fails. Your agent writes a failing test, builds the change, fixes failures and commits. It may delegate independent slices.
Open full size ↗
Open full size ↗
- Test runs are recorded with their commit and exit code, so Verify can cite them.
- Amendments appear under Decided while building, each with whether it changes what you approved. A change to what you approved becomes a visible amendment.
- A login, secret or other human-only prerequisite never counts as passed. Other work continues, you get one question at the end of Verify, and the App never takes the secret.
- Outline and Implement ask no workflow questions; Verify asks at most one, at its end.
- Runs start in the agent's own permission mode. Native permission and sign-in prompts still need your response, and an explicit stop or pause stays in force.
When the build is done
The session decides when the build is ready, and Verify starts on what it committed. Unticked cards and open needs stay open in Verify, never passed.