Repository navigation
fix(conductor): skip the in-chat task board when the Dashboard tab shows it - #18608
Conversation
UX Review (Opus 5.5) — ⏭️ skippedRevision |
First Principles Review (Opus 5.5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS Check before merge: the description says decision #18900 "replaces" What this change shipsInventory (5 items) — 5 justifiedIntent: stop a conductor whose Dashboard tab already shows the task board from redrawing that board in chat at every milestone. FIX.
[FIRST-PRINCIPLES-REVIEWED] 4f31506 Description read[DESCRIPTION-READ] a5579f3ac867938c06ceaac5a45c5acce3d57bb7c22c654f6202cf41bcd4a8bd That sha256 names the PR description this verdict read. Recompute it with |
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: |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Check one live [DESIGN-REVIEWED] 4f31506 Description read[DESCRIPTION-READ] 9e0440859d37afba99007b9226f97d622556824a3121ac71881dcd8bb9438621 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 |
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: |
|
Parked while we wait for @bolichen97's on-record decision. Details and the reason are on the RFC: #18620. The recorded-decisions and First Principles blocks clear only after his reply, so there will be no push here until then. |
f566336 to
6bd42fd
Compare
6bd42fd to
dfc0e9c
Compare
dfc0e9c to
ae8d6df
Compare
17b390b to
62d2f56
Compare
…ows it A crewmate conductor whose Dashboard tab shows `goal-board` or `work-kanban` already has the full task board beside the chat, kept current from the work ledger. Drawing the same board as a widget at every milestone showed every task twice. The "Talking to the person" milestone rule now branches on the `[DASHBOARD]` block. With one of those two templates in force, the reply carries only the "Needs you" lines (when anything waits on the person) and one short line on what changed and what happens next. Every other surface keeps the full "Task board" widget, and the `task-dashboard` artifact rule stays with it, since those surfaces have no tab carrying the full board. The intro widget rule is unchanged. Pinned conductor spec digests and prompt fingerprint are refreshed.
62d2f56 to
4f31506
Compare
KiroCrewCaptain
left a comment
There was a problem hiding this comment.
Tier 1 auto-approve: fix (3 files). Criteria: no conflict, no requested changes, security path denylist clean, design-doc gate clean, security checklist all-NO, no reviewer-lane findings, readiness signal fresh for this head SHA, declared category matches the diff. Category: conductor prompt stops drawing the in-chat task board when the Dashboard tab already shows a board page; spec hash fixtures updated to match.
Problem / Motivation
Goal: A Crew Mode conductor whose Dashboard tab already shows the task board stops drawing a second full copy of that board in the chat at every milestone.
User feedback (Crew Mode, dashboard chat): at every milestone the conductor draws a full "Task board" widget in the chat. The conductor's Dashboard tab, on the
goal-boardtemplate, already shows the same board, folded live from the work ledger. Every task is shown twice, and the chat fills with tables the reader already has beside it.Why it matters
Crew Mode is the plain-words entry to the conductor for everyday users. A chat where each milestone repeats a full table that sits one tab away buries the one thing they need to read: what waits on them.
Not a goal
goal-board/work-kanbantemplates or the[DASHBOARD]block.What changed (motivation → approach → change)
Symptom (board twice) → cause: the "Talking to the person" milestone rule in
_CONDUCTOR_SYSTEM_PROMPT(src/kiro_crew/agent.py) asks for the full widget unconditionally, though the conductor's turn already carries a[DASHBOARD]block naming the template its tab shows → the rule now branches on that block.This is a generation-time rule in the conductor prompt: on a board tab the model is told not to write the widget or the artifact, so nothing is drawn and then hidden, and no client-side suppression is added. The motivation is the reader's duplication AND the output tokens of a full widget plus a
task-dashboardartifact update at every milestone. Design record: #18620.goal-boardorwork-kanban(the two templates that draw every work item from theworkfold): the reply carries only the "Needs you" lines first when anything waits on the person, then one short line on what changed and what happens next. No per-task table.[DASHBOARD]block exists", because a crewmate that never adopted a template gets the defaultproject-report, which does not draw the task board.task-dashboardartifact rule: written on the no-tab branch only. There it is still the only full board. On a board tab the prompt says not to write it: the tab is the full board, and a third copy costs output tokens.[OPTIONS: ...]/ask_questionrule for answers now sits above both branches, since both still ask the person things.Recorded decision: RFC #18620 (
docs/request-for-change/rfc-conductor-chat-board-on-tab.md) is merged on main by maintainer @iamwhatever. The superseding decision entrydocs/decisions/2026-10-10-conductor-board-tab-replaces-the-in-chat-board.md, which replaces2026-10-02-conductor-introduces-itself-in-every-chat.md, is in its own PR #18900, asdocs/decisions/README.mdrule 3 asks.@bolichen97 — this changes the Crew Mode milestone rule you added in #16088. Could you review whether this split matches your intent?
Backwards compatibility
Compatible: prompt text only; no API, config or data shape changes.
The diff takes one thing away on one surface: the in-chat per-task table when the Dashboard tab shows
goal-board/work-kanban.Reader: src/kiro_crew/agent.py:_CONDUCTOR_SYSTEM_PROMPT -- chat -- the conductor model reads it; the same board stays on its Dashboard tab (
dashboard_templates/builtin/goal-board), and "Needs you" still leads the chat replyReader: src/kiro_crew/dashboard_agentic.py:turn_block -- crew page -- unchanged; it is the signal the new branch reads
Reader: test/test_agent_refactor_spec_bytes.py:GOLDEN -- release -- digests refreshed for the new prompt bytes (test_every_written_spec_matches_the_pre_split_bytes)
Tests
test/test_agent_refactor_spec_bytes.py: conductor and ledger-conductor spec digests refreshed for the new prompt bytes (all eight scenarios).test/test_agent_refactor_contracts.py:_CONDUCTOR_SYSTEM_PROMPTfingerprint refreshed.test_agent_refactor_facade,test_conductor_agent,test_conductor_base_work_ledger,test_ledger_conductor_agent,test_conductor_allowed_tools_survive,test_conductor_wake_reset,test_agent_execution_catalog368 passed. black / isort / flake8 clean on the touched files.Manual verification
N/A — prompt-only change; no code path reads the text beyond the pinned digests.
Related Issues
Follows #16088. RFC #18620 (merged). Superseding decision #18900.
Pattern harvest
Not generalizable: one prompt rule written before the conductor's Dashboard tab carried the same board; nothing else draws a second copy of a tab's content.
Checklist
Contribution License Agreement
Contributions are accepted under the project's Apache-2.0 license.