Agent Runs & the Plan Gate
A structured execution of an AI agent against one or more intents. The Plan Gate is the critical human checkpoint before execution begins.
The Run Board
Every run in the studio is on the Run Board, newest first, each with a link back to what started it — a Work Package, an intent, or a repository scan. Filter by what needs you, by agent, by initiator or by date range, and sort by when the run began. Coding runs for Work Packages appear alongside agent runs, with their branch and pull request, and Outstanding pull requests lists every open pull request with its package, its Definition of Done state and its merge hold. See Delivery: Coding Runs, Pull Requests & the Merge Hold.
Creating a Run
From the Run Board or intent detail, click New Run. Select the intent(s), agent type, and context overrides. The run enters Queued state.
The Plan Gate
The agent generates a step-by-step plan. Read every step: Does this match the intent? Are there steps touching things they should not? Is the estimated cost acceptable? Acknowledge to proceed, or reject to request a revised plan.
Run Status Reference
| Status | Meaning | Action required |
|---|---|---|
| Queued | Waiting for agent capacity | None |
| Planning | Agent generating plan | Wait for notification |
| Awaiting Plan Approval | Plan ready for review | Acknowledge or reject |
| Running | Agent executing | Monitor; answer questions promptly |
| Awaiting Review | Output ready (runs started before the Definition of Done read pull requests) | Accept or reject under Earlier runs on the Review feed |
| Complete | Run finished | Its work package’s Definition of Done decides the rest |
| Failed | Unrecoverable error | Review error log; re-run |