Briefs & states
A brief is a single piece of work — a bug, a feature request, a customer note, a task. Every brief moves through the same six-state lifecycle. Two states are terminal, everything else can go forward or back, and nothing ever ends up at a dead end.
The lifecycle
Below is the full set of legal moves. Anything not pictured is rejected by the API and the UI alike, so you can't accidentally skip the review step or push a brief from done back into the active queue.
Just landed. The flagging agent is about to look at it; humans can also triage from here.
Refined and assigned. Either a human picks it up, or the coding agent starts work.
Someone is on it — usually the coding agent. The brief stays locked to that owner so nothing overlaps.
There is a PR or a draft ready. The reviewer either approves or sends it back with feedback.
The reviewer signed off. Counted in stats, hidden from active lists.
Junk, spam, duplicate, or a deliberate close. Counted separately from done.
What you'll see on a brief
The brief detail page is organised around the things you actually act on:
- Title and description — what the user said, then what the refining agent rewrote it as. The agent never overwrites your edits.
- Type — bug, feature request, customer feedback, enquiry, task, or internal note.
- Priority — low, medium, high, critical.
- Assignee — a teammate or one of the agents (assignment is what gets things moving, not state).
- Confidence — how sure the refining agent is of its rewrite. Used by workflow rules to decide whether to auto-promote.
- Labels — free-form tags, AI-suggested or human-added.
- Attachments — screenshots, logs, anything the submitter sent along.
- Timeline — every state change and every agent run, with who did what and when.
- Attempt history — when a brief gets reopened, the previous PR, the agent's summary and the reviewer's feedback all stay on the brief.
Reopen with feedback
From review, click Reopen with feedback, type what should change, and the brief flips back to ready. The previous attempt isn't lost — its branch, the agent's summary, and your feedback all stay attached. The next time the coding agent picks the brief up, it starts already knowing what was tried, what shipped, and what you didn't like.
Each new attempt gets its own branch. Old work is never overwritten, and you can always go back to compare.
No dead ends. Even cancelled isn't really the end — a cancelled brief stays searchable, with everything that was submitted, scored, or said about it. If the bouncer got it wrong, you can pull it back.