Repository navigation
fix(crew): draw the conductor's drawer board from its work ledger - #18957
iamwhatever wants to merge 2 commits into
Conversation
First Principles Review (Opus 5.5) — 🟡 CONCERNSPremise-level review of First-Principles-Verdict: CONCERNS A new 60 s poll and a new Not justified as shipped
What this change shipsInventory (8 items) — 5 justifiedIntent: keep the conductor's drawer board current without the conductor having to republish it. FIX (#18956).
[FIRST-PRINCIPLES-REVIEWED] 45c576d Description read[DESCRIPTION-READ] 5984b3db006ac73d001a6c63e94873a7aa308d2b8bd264a992afa7e734fe2f0e That sha256 names the PR description this verdict read. Recompute it with |
UX Review (Opus 5.5) — 🟡 CONCERNSUX-level review of UX-Verdict: CONCERNS The muted "Last plan" line still tells the reader to wait on a task the board shows as done, so the board seems to contradict itself. Watch
Suggestions
[UX-REVIEWED] 45c576d Description read[DESCRIPTION-READ] 60fd5c7283c0f1c058009d75459bb062cfbe0114c3e75bb5069c20f01f6805b7 That sha256 names everything this verdict read: the PR description, plus the 5 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 |
Design Review (Opus 5.5) — 🟡 CONCERNSDesign-level review of Design-Verdict: CONCERNS The skill now says "do not republish", yet a slot shared by two crews still serves only what was published, so that conductor's board never updates. Watch
Suggestions
[DESIGN-REVIEWED] 45c576d Description read[DESCRIPTION-READ] 494f76f5dc3b015c331639aedd30a26b454f789658b11821f076b1c442946220 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 |
Opus 5.5 Review — ✅ no blocking findingsReviewed Review detailsNo findings that block the merge. FINDING — docs/system-specs/modules/learn-cron-dashboard.md:2552 — [OPUS-REVIEWED] 45c576d Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
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: |
7740491 to
29b9cd5
Compare
29b9cd5 to
259293c
Compare
259293c to
5facc97
Compare
5facc97 to
84d6b6d
Compare
|
Dispositions for the CONCERNS on 893742a (author reply): Design: wrong crews get the conductor board — fixed in 05e4d1d. Verified in code: First Principles: duplicate 60 s constant — accepted, not changed here. UX: stale "Next (from )" label — accepted as a wording follow-up. The label already carries the time it was written, which is the honest part; making it say "may be out of date" is copy polish that does not change what the board shows. Left for a follow-up so this PR does not take another full review round. |
be4cf20 to
b17ac58
Compare
|
Self-review (crew-pr-reviewer, read-only): CHANGES NEEDED This PR rebuilds the goal conductor's drawer board from the work ledger on every read, so a skipped |
|
Converted to draft by crew-pr-reviewer: the self-review above is CHANGES NEEDED. Mark ready again (gh pr ready 18957) once the points are addressed. |
b17ac58 to
5965892
Compare
|
Addressed the crew-pr-reviewer self-review and the UX CONCERNS on b17ac58, on head 5965892 (rebased onto main 0cda152, still two commits). Self-review
UX CONCERNS (b17ac58)
Verified locally: touched pytest ( |
5965892 to
54d3d3e
Compare
|
Disposition for UX Review CONCERNS on 54d3d3e (advisory, no push):
Remaining CI red (Backend Tests shard 2, Coverage Gate) is |
54d3d3e to
4272586
Compare
|
Disposition for UX Review CONCERNS on f1ab384 (advisory, no push): these are the same two points answered on 54d3d3e in #18957 (comment), and that answer stands. The matching minutes come from the capture fixture writing the publish and the ledger entries a second apart; in use |
The conductor's right-side panel (Dashboard preview off) showed only what the agent last sent through panel_publish, so a cycle that skipped the publish left new work items off the board for hours while the work ledger already held them. The drawer read now rebuilds the kirocrew-conductor board from the same work fold goal-board reads: tasks, state, PR, the done count and the needs-you band (an open item whose worker reported question or blocked). A crewmate with a work ledger that never published gets that board too, on a slot that names it alone. Published data adds only `next` and `repo`; a `next` older than the ledger's newest entry carries `next_from` and renders "Next (from <time>)". The drawer refetches on member_projection and a 60 s foreground poll, the same liveness the Dynamic Dashboard tab keeps. The goal-conductor skill no longer asks for a republish each cycle.
The no-publish path built a conductor board for any crew that is not the pipeline conductor, and every crew carries a work ledger. The gate now names the conductor template positively, a test pins that another crew with a ledger and no publish still gets no board, and the feature map names the conductor crew instead of any crewmate. A queued item that no worker has started reports no step, so the board lights no segment for it; only a started item sits on Code. The drawer's 60 s re-read runs only for the conductor's board (by template, or by slug before any board exists), sharing the Dashboard tab's DASHBOARD_FALLBACK_REFETCH_MS from a leaf module so the lazy tab stays out of the drawer's chunk. When the published Next line is older than the ledger, the board shows it muted as "Last plan (time):" so it no longer reads as the current instruction.
|
Closing as superseded. #19659 (merged to kirodotdev/KiroCrew main on 2026-10-13, commit 1f6f5af) removed the Dynamic Dashboard preview flag and deleted |
Problem / Motivation
Goal: The goal conductor's side-panel task board stays current on its own, without the conductor having to remember to republish it.
The crew page's side panel (Dashboard preview off) shows only what the conductor last sent with
panel_publish. When a patrol cycle skips the publish, the board goes stale for hours: new items never show up and finished ones keep their old state. The work ledger already holds every item the whole time.Why it matters
The board is where a person looks to see what needs them. A stale board hides new work and new questions until the person asks the conductor in chat. Asking the agent to "remember to update" does not fix that, because forgetting is the failure.
Not a goal
goal-board,CrewDynamicDashboard); it already reads the ledger.What changed (motivation → approach → change)
question/blockednextlineLast plan (<time>)once the ledger moved onpanel_publishedonlyThe board was hand-written each cycle, so a skipped publish was a stale board. The drawer read (
dashboard/handlers/agent_panel.py,_panel_record) already rebuilds the pipeline conductor's board from theworkfold. It now does the same forkirocrew-conductorthrough the new providerbuild_conductor_boardinconductor_board_contract.py. Published data is read only fornextandrepo. The ledger board is built only when the slot names this crew alone and a configured crew of that exact name owns the slug (otherwise the board is served as published), and its worker-written text goes through the same redaction a publish gets (src/kiro_crew/agent_panel.py:scrub_data,derived_record).The template
agent_panel_templates/kirocrew-conductor.htmlshows an oldnextmuted, asLast plan (<next_from>). A queued item has an empty step, so its bar lights nothing. The drawer (pages/members/CrewWebview.tsx) re-reads the conductor's board on a 60 s foreground poll, because a ledger entry sends no frame that names this query; the poll keys on the record's template or the conductor's slug (pages/members/dashboardRefetch.ts) and reuses the Dynamic Dashboard tab'sDASHBOARD_FALLBACK_REFETCH_MS. Every other drawer stays push-only. Thegoal-conductorskill (builtin_skills/goal-conductor/SKILL.md) no longer asks for a republish each cycle. Docs updated indocs/feature-map/README.md.Backwards compatibility
Compatible:
panel_publishaccepts the same data; a published board is still served when no work ledger exists, and every other template is served exactly as published.Removes nothing: the published task list is no longer the source of the conductor board, but it is still stored and read; only the fields the ledger owns are replaced on read.
Tests
test/test_conductor_board_from_fold.py(new): an item created in the ledger appears on the drawer with nopanel_publish; items created after the last publish appear without another; needs-you comes from a worker'squestion; a stalenextgetsnext_from; a board with no ledger is served as published; nothing is synthesized without a ledger or on a shared slot.origin/mainthe three ledger-to-drawer tests fail (assert None is not None, the old one-item list); all six pass here.test/test_agent_panel_routes.py: the single-read counter stub forwards the new keyword.test_every_item_the_fold_kept_is_on_the_board: the task list and the done count cover the same items.test_a_published_board_on_a_shared_slot_never_takes_the_ledgers_tasksandtest_a_ledger_board_is_served_only_to_the_crew_that_owns_the_slug: another crew's ledger never reaches this crew's drawer, and a caller-chosenmembername gets an empty panel.Manual verification
Rendered the real drawer document through
_panel_recordagainst a real crew log (one publish, then three ledger items, one asking a question) and captured it at 420 px, before and after.Screenshots / video
Before (main): the drawer shows only the last publish, one task, nothing needs you.
After (this head): three tasks from the ledger, U's question in Needs you, 1 of 3 done, the old
nextmuted asLast plan (…), and queued V with no step lit.Related Issues
Closes #18956
Refs #18919
Pattern harvest
Rule candidate: review-prompt
Pattern: a UI surface whose state an agent must re-send each cycle, while a log fold already holds that state.
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)Contribution License Agreement
Contributions are accepted under the project's Apache-2.0 license.