Repository navigation
fix: chat header status dot reflects the stop phase - #21501
Conversation
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Check that the two headers agree when no turn is running: the ChatPane dot needs [DESIGN-REVIEWED] c4e35ae Description read[DESCRIPTION-READ] b74ac36a7a09f0c6ee8cb6aea6868c69941f3a0c328b6f41fbe3c08bd1ce17f5 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 Check one thing: the ChatPane header caption now shows a localized "Stopping…"/"Killing…" in place of the raw run-state token, and the Goal names only the dot. What this change shipsInventory (6 items) — 6 justifiedIntent: A user watching a chat header sees one consistent signal that their stop is under way, not a green "working" dot. This is a FIX.
[FIRST-PRINCIPLES-REVIEWED] c4e35ae Description read[DESCRIPTION-READ] e71af3ba7069899b0818bb6e2b41d91dc87af9790a332162e05853be06b1ea41 That sha256 names the PR description this verdict read. Recompute it with |
UX Review (Opus 5.5) — ✅ PASSUX-level review of UX-Verdict: PASS Check whether a pulsing red dot for a stop the user started themselves reads as an error. The blind reader called it "a warning or 'shutting down' light." [UX-REVIEWED] c4e35ae Description read[DESCRIPTION-READ] 7e4ec059c8849eda3211a5027ead8aa1b0f62b1ea777d48d87e4acb04fcd72b9 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: |
7a71f88 to
c7a9aff
Compare
c7a9aff to
c657601
Compare
afc0c67 to
04518c3
Compare
fa2ac04 to
b12c359
Compare
|
Rebased: main moved in website/src/components (same-dir as this PR's files) since this head went green; CI re-running. |
|
self-added: yes
|
|
Frontend Tests (2) red is inherited base breakage, not this PR. The only failing test is Root cause and fix are already tracked: #21535 (domGolden katex snapshot stale after katex 0.18.2 class rename) and PR #21536. This PR will rebase onto that fix once it lands and re-run to green; no change here is warranted. |
|
Update: the red checks have grown to |
b12c359 to
c4e35ae
Compare
|
Rebased onto current |
KiroCrewCaptain
left a comment
There was a problem hiding this comment.
Tier 1 auto-approve: fix (6 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: display-only fix that routes the chat header status dot (ChatPane and ChatEmbed) through the existing stop-under-way signal so it turns red during a stop, with a regression test and a capture harness.
































Problem / Motivation
Goal: Make the chat header's status dot turn red while a stop is under way, so it agrees with the "Stopping…" text instead of staying green as if the turn were working normally.
While a stop is in flight, the chat header's status dot stayed green — the same colour it shows for a turn that is working normally — so the dot and the "Stopping…" caption beside it disagreed. UX review on #19898 flagged it. The main chat header (the shared ChatPane header used by split-view panes and Crew Member DM panes) and the app-sdk ChatEmbed header both followed the same rule, so the mismatch was shared, not embed-only.
Why it matters
The status dot is the one glanceable signal of what a session is doing. A green dot means "working"; leaving it green through a stop tells the user the opposite of what the header text, the footer and the composer all say, so a stop in progress reads as a healthy running turn. One session can only be in one state, and the dot must say which.
Not a goal
--ok/--dangertokens the footer caption already uses.ChatTitleOverlay), which has no status dot.What changed (motivation → approach → change)
The two chat-header surfaces coloured their dot
bg-ok(green) whenever the turn was running, with no branch for the stop phase — so a stop in flight kept the working colour. Each surface already computed the stop-under-way signal for the text beside the dot; the dot simply did not read it.The fix routes the dot through that same signal. In
ChatPane.tsxa singlestopUnderWayis derived from exactly what its footer reads (streamState === 'stopping' || slots-frame stopping) and now feeds the dot, the header caption, the footerstoppingprop and the composerisQueuedprop — one copy where three inline copies stood. InChatEmbed.tsxthe dot reuses the existingstopUnderWay(stop_state !== 'idle' || slots-frame stopping). When it holds, the dot isbg-danger animate-pulse; otherwise the prior green-while-running / accent-while-idle (and muted-while-stalled, embed only) behaviour is unchanged.ChatPane.tsx's header caption also follows: while a stop is under way it reads a localized "Stopping…" intext-dangerinstead of the green run-state token, so the dot and the caption right beside it no longer contradict each other. Header, caption, footer and composer now describe one moment.Backwards compatibility
Compatible: a className-only change on two existing header dots, gated on signals already in the store/poll; no contract, prop, API or data shape changes.
Removes nothing: no path, record, validator, key or function is hidden, removed or tightened.
Tests
website/src/test/chatHeaderStopDot.test.tsx(new): renders the REALChatPane(non-frameless, so the header dot renders) and the REALChatEmbed, and asserts the dot isbg-ok(notbg-danger) while a turn works normally andbg-danger(notbg-ok) while a stop is under way — forChatPaneboth via thestoppingstream state and via the slots-framestoppingflag, and forChatEmbedvia the polledstop_state. It also asserts theChatPaneheader caption reads the green run-state token while working and a danger "Stopping…" while stopping. Proven: the "turns danger" cases fail on the parent commit (dot staysbg-ok) and pass on the fix; the "is green" cases pass on both.Manual verification
Captured with the committed Playwright harness
website/scripts/capture-20397-chat-header-stop-dot.mjs, which asserts the dot's computed background equals--okwhile working and--dangerwhile stopping (and that the two differ) before writing each PNG. See Screenshots.Screenshots / video
Both header surfaces, dark and light, across all three states: working (green dot, run-state caption), a soft stop ("Stopping…"), and a force kill ("Killing…"). The dot uses the SAME
--dangerthe footer and composer already use for the stop phase, and the caption reads the SAME localized "Stopping…"/"Killing…" the footer and ChatEmbed show — so the header reads as the stop phase the user asked for, in step with the rest of the surface, not as a new error state.ChatPane header (split view / Crew Member DM) — working, stopping, killing (dark):
ChatEmbed header (app-sdk) — working, stopping, killing (dark):
Light theme (same three states, both surfaces)
ChatPane — working, stopping, killing:
ChatEmbed — working, stopping, killing:
Related Issues
Closes #20397
Pattern harvest
Not generalizable: a two-site display fix where the status indicator read a narrower signal (running) than the sibling text it sits with (stop phase); the fix points both at the one signal. No broader class to ratchet.
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)