Repository navigation
feat(dashboard): say when a conductor mode run is done or stopped - #37058
Conversation
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Nothing blocks: the verdict is computed on each read from facts the gateway already holds, nothing new is stored, and the field is additive. Notes
[DESIGN-REVIEWED] 47d3e1f Description read[DESCRIPTION-READ] 7f5cf1eda2bca467a9d7dc1cca03c9cb53e6d156662c28050c4977831fee83d4 That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
First Principles Review (Opus 5.5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS Nothing blocks; the Notes
What this change shipsInventory (7 items) — 6 justifiedIntent: a person watching a conductor mode chat can tell whether its run is done, stopped or still working. This is an ADDITION, and it also fixes two places where the dashboard claimed more than was true (#36912).
[FIRST-PRINCIPLES-REVIEWED] 47d3e1f Description read[DESCRIPTION-READ] 400b5c363e368173161e08903de7c3095abd2db0ae0ccfe426ae1e8fa0745ce9 That sha256 names the PR description this verdict read. Recompute it with |
UX Review (Opus 5.5) — ✅ PASSUX-level review of UX-Verdict: PASS Nothing blocks. The blind reader read all five run lines and both "Not done" rows the way the diff means them, and every state appears in a screenshot. Notes
[UX-REVIEWED] 47d3e1f Description read[DESCRIPTION-READ] bbe48022bcb64dd3a0660d6f4eb1e4cc5313f21cec59df7e25f90076efe9531c That sha256 names everything this verdict read: the PR description, plus the 15 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
GPT 6.1 Review — ✅ no blocking findingsGPT 6.1 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
Opus 5.5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
Sensitive Change Review
|
493b884 to
265776b
Compare
|
|
Code change approver recommendation: A person should decideWhy: the decider estimates a 12% chance that this change introduces a defect, within the 12% limit for approving on its own, but it changes 970 lines, more than the 400 it may approve on its own. Additive read-only An automated recommendation, not a review: a maintainer still decides. It approves on its own only when the estimated chance of a defect is at most 12% and the change has at most 400 changed lines. Decider v3c (217fe607) · commit |
265776b to
3409902
Compare
|
|
ff3d1ab to
db80d1a
Compare
|
db80d1a to
606bbaf
Compare
|
|
606bbaf to
58e13e3
Compare
58e13e3 to
261c176
Compare
|
261c176 to
a5ecb46
Compare
|
Two reds on
The previous head |
The gateway derives a settled verdict from the work ledger item states and whether a loop is still active on the conductor session: done when every item is closed, stopped when some are open. GET /api/crew-board and the root row crew_work carry it as run. The Board Overview opens with one run line, a rejected or abandoned milestone reads Not done, and the Sessions row says Done only when the run is done. Closes #36912
a5ecb46 to
47d3e1f
Compare
KiroCrewCaptain
left a comment
There was a problem hiding this comment.
Approved. On head 47d3e1f the review read the whole diff and checked the callers of each changed function. It confirmed that patrol_of only runs on the event loop and that a missing nudge service reads as off. It also checked that the new _closed cache is cleared beside every _items write, and that the take-away changes to milestoneState and the Sessions row allDone have no readers outside this diff. In a scratch copy the backend tests passed 24 of 24, and the three related files passed 59 of 59 three times in a row. A mutation run showed that test_a_closed_count_change_alone_still_moves_the_row fails when the change check in crew_work.remember is removed.
Done well: "done" comes from facts the gateway already holds, so it stores nothing that can go stale. A rejected or abandoned milestone also no longer shows the green check.
- Optional: no test covers the legacy header branch in
work_ledger_board.py:801. Replacing it withgoal_items=record.goal_items_usedfails 0 of 113 tests. One board read whoseconductor.jsonhas nogoal_items_basekey would pin it. - Optional: nothing outside tests reads
run.openorrun.accepted. You could drop them, or give the First-Principles note a disposition that says why they stay. - Optional:
patrol_ofcounts any active loop on the slot, but the siblingconductor_loopingcounts only the work-ledger patrol. Use one definition, or say in work-ledger.md that any watch keeps the run Working.
Problem / Motivation
Goal: A person looking at a conductor mode chat can tell when its run is done, stopped or still working.
Today the Board never says a run is finished. After every milestone is accepted it still reads "Done when ...", which looks pending. A rejected or abandoned milestone shows the green done check and counts toward "X of Y". A run that stopped with milestones open looks stalled or stuck. The Sessions row says "Done" when every worker session finished, without reading the work ledger, so a run stopped with open work can read "Done".
Why it matters
Conductor mode runs a goal through many worker sessions over hours. If the person cannot see that the run ended, and whether it ended with everything accepted or with work left, they either keep waiting on a finished run or miss work that was dropped.
Not a goal
What changed (motivation → approach → change)
A conductor already ends its run by sending one final report and then stopping its patrol loop. So the gateway can tell a finished run from facts it already holds, with nothing new stored:
dashboard/conductor_mode_run.pyworks this out asrun(settled,open,accepted,not_done,at_limit,patrol_on).GET /api/crew-boardcarries it, and keeps a new goal with no item yet unsettled through the header'sgoal_items_used. The root row'screw_workcarries it once its board has an item; the row cannot see the goal's start, which the spec lists as a known limit.lib/conductorModeRun.tsputs what is live first). An unsettled board reads Working only while the patrol is on, so a new goal waiting for its go-ahead with no milestone yet draws no line. Done and Stopped add one plain sentence with no numbers. The line adds no control; the panel's Back row stays the one way back. No line claims a final report exists, because a stop at a limit or by a person sends none. Done with milestones left keeps a muted check; Stopped shows a stop sign.run.settledisdone; a root with no ledger keeps the every-worker-done rule.website/src/i18n(English inlocales/en.manual.json), therunfield on the board and slot types (types/index.chat.ts,api/crewBoard.ts), a capture page for the five states, and the rule written down in the crew-mode spec anddocs/work-ledger.md.The alternative, a "finished" record the conductor writes, needed a ledger field and prompt text in four places and could go stale after a new goal, so it was not taken.
Backwards compatibility
Compatible:
runis a new optional field on two reads; an older dashboard ignores it, and this dashboard keeps today's behaviour when it is absent.Removes nothing: no field, key, record or path is removed. The row's "Done" and the milestone check now follow the ledger, which corrects what they claimed.
Tests
test/test_conductor_mode_run.py: the verdict table, a new goal with no item yet staying unsettled on the board read, the loop reading (active, stopped, budget stop, no loop),patrol_onoff for a new goal waiting on its first item, the row'srunwith the fold's forgotten closes, norunfor an empty board, and the board read'srun.website/src/lib/conductorModeRun.test.ts: the state order.website/src/pages/chat/command-center/ConductorModeCrewPanel.run.test.tsx: the Overview line in each state, Not done rows, no second way back beside the panel's Back row, one notice per failed reopen, Waiting for you while the main chat holds its own tool approval, no claim of a final report (a stop at a limit included), no line without a goal.website/src/test/ChatSidebar.conductorLane.test.tsx: Done followsrun.settled; a stopped run keeps the usual line.ConductorModeCrewPanel.test.ts: an abandoned or rejected item isnot_done.The new run-line and row tests fail on the base commit, except the no-goal test, which pins today's behaviour. The backend tests passed 20 times at
-n0, 10 times at-n 4besidetest_crew_work_summary.pyandtest_crew_board_asks.py, and once in a shuffled order.The final head is a content-identical rebase of the head both local review lanes passed (
db5581004f):git diff <base>..<head> | git patch-id --stablegives394a03a56ea8fe59a951fef974e0235fc138a8f5for both.Manual verification
Rendered the real Board panel from
website/capture/conductor-mode-run-states.htmlin Chromium, before (base) and after, for each state, and looked at every image.Screenshots / video
Before → after, Board Status view at 360 px.
Working, Waiting for you and Paused
Related Issues
Closes #36912
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)