Follow the Oratorio workflow
Use the Board to move a task from intake through Agent work, review, and a recorded decision.
Find work on the Board
The Active view groups current tasks by lifecycle stage. Use All, Cancelled, and Archived for completed or inactive work.
- Search titles and descriptions, or narrow results with
source:andlabel:qualifiers. - Filter by repository or assignee.
- Select Sync sources to request an immediate GitHub and GitLab update.
- Select a card to open Quick View without losing the current Board filters.
Quick View shows the current status, recent activity, drafts, comments, and only the actions currently allowed by Oratorio.
Start and follow Agent work
Open a discovered task and choose one of the offered run modes. The available choices depend on the task source and its current state.
Each run uses an Oratorio-managed worktree. Open the linked DotCraft thread when you need the full conversation, plan, tool activity, or file changes. Cancelling an active run requires confirmation.
Review the task detail
Task Detail keeps the workflow in five stages:
- Intake — problem statement, source metadata, labels, assignee, and base branch.
- Analysis — run attempts, live activity, timeline, diagnostics, and worktree information.
- Review — Agent drafts, findings, suggestions, comments, implementation delivery, and follow-up tasks.
- Decision — approve, request changes, or reject, plus the status of any provider write.
- Closed — the recorded outcome, history, archive, reopen, and re-review actions when available.
Deliver or continue the work
Review drafts can contain inline findings and suggested replacements. Resolve or reopen findings before publishing when more discussion is needed.
Implementation drafts can deliver a branch as a GitHub pull request or GitLab merge request when source writes are enabled. Follow-up drafts can be edited and turned into new local tasks. If a provider write fails, Oratorio keeps the recorded decision and offers a targeted retry.
Oratorio refreshes the task and Board after successful commands. A reconnect refreshes state but does not repeat the original user action.