- Plan
- Review planRevise and approve
- Build
- TestEarly suite off by default
- Review codeFix and review again if needed
- LandRebase → retest → land
- Closeout
Needs your input holds a decision at the stage that asked. Answer there to continue.
Plan
The planner reads the request and prepares an implementation plan.
What the app shows: Planning.
What can stop it here: Too big means the request has more than one independently deliverable increment or more than three acceptance sequences. Read the proposal and split or reduce the request.
Review the plan
A separate reviewer checks the plan. The fold step revises it with that review. Approval then follows the pinned policy: a person decides, or the attendant judges whether it can proceed.
What the app shows: Reviewing the plan, then Revising the plan. A decision for you appears in Needs your input.
What can stop it here: A plan decision or product question needs an answer. Open the question and read the plan before responding.
Build
The builder implements the accepted plan and runs the tests its task touches.
What the app shows: Building.
What can stop it here: A tool failure, a question or an attempt limit can interrupt work. Read the recorded reason before restarting or granting more attempts.
Test
The line has an early test row and a retest row after rebase. In the shipped line, early testing is disabled: build goes to code review, and the suite runs at retest before landing.
What the app shows: Testing when the early test row is enabled. Retesting when the suite runs after rebase.
What can stop it here: A failing check needs investigation. Base red identifies failures on the base.
Review the code
Code review checks the implementation. Findings can route it through a fix and back to review. A clean review proceeds to rebase and retest.
What the app shows: Reviewing. Fixing when review sends work back.
What can stop it here: Unresolved findings can reach the attempt cap. Read the findings before deciding whether more attempts are useful.
Land
The driver rebases the change, runs the required checks and lands it on the configured target. A competing change can require another rebase and retest.
What the app shows: Rebasing, Retesting and Landing.
What can stop it here: Conflicts, failed checks or a failing base can prevent landing. Follow the recorded reason; extra attempts do not repair the base.
Closeout
Closeout records the result and performs the configured tracker transition. A local request completes without a tracker.
What the app shows: Closing out, then Done after successful closeout.
What can stop it here: A failed closeout needs the recorded failure resolved. A stopped ticket does not change the tracker.
Reading an open ticket
An open ticket has four tabs. Overview shows the stage, its status line and the reason for a stop. Plan shows the accepted plan and its review. Agent log shows what each session did, stage by stage: its messages, tool calls and results, and the files it changed. Changes shows the commits and diffs grouped by the task that produced them.
What the app shows: Overview, Plan, Agent log and Changes on the ticket page, with an Activity panel beside them.
Where Needs your input fits
Needs your input is a place to answer decisions, not an extra pipeline stage. Open the ticket there and read what it is asking: approve a plan, answer a product question, or decide whether to grant more attempts. Work waits at the stage that raised it.
What the app shows: Needs your input. When the attendant leaves a decision to you, its context appears with the question.
What can stop it here: An unanswered decision stays with you. If the reason is unclear, use Report this action on the ticket rather than guessing.