Skip to content

feat: crewmate chat shows a typing bubble while the crewmate works - #20032

Merged
iamwhatever merged 1 commit into
mainfrom
feat/crewmate-typing-bubble
Oct 14, 2026
Merged

iamwhatever merged 1 commit into
mainfrom
feat/crewmate-typing-bubble

Conversation

@CrysisDeu

@CrysisDeu CrysisDeu commented Oct 14, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

Goal: While a crewmate works, its DM thread shows a typing bubble — the way a messaging app shows a reply being typed — instead of a loading indicator.

Today a crewmate's DM thread (Members page, ChatPane) shows two things under the last message while the crewmate's turn runs: a grey one-line step label (CrewmateLiveActivity) and the main chat page's ghost-pose working carousel (ChatFooter). Together they read as loading, not as a conversation. The product frames a crewmate DM as messaging a teammate; the reply-in-progress should look like one.

Why it matters

The crewmate DM is the surface where a user watches a crewmate work in real time. A spinner-shaped indicator makes it feel like a tool, and two moving things (carousel + swapping text) compete for attention. A typing bubble is the pattern every messaging user already knows and reads at zero context.

Not a goal

  • Changing the main chat page's working indicator (the ghost-pose carousel stays, theme-swappable as before).
  • Changing what the caption says: its words are the DM header pill's own reading (useSlotActivity, toolStatusLabel); rewording tool titles or "Compacting…" is a separate change to that shared seam.
  • Changing the hold timings (useHeldActivity).

What changed (motivation → approach → change)

Goal → a bubble in the crewmate's column that reads as "a reply is being typed" → three mockups were reviewed (dots only / dots + caption below / dots + caption inside one bubble); "dots + caption below" was chosen because it keeps the what-is-it-doing signal without letting the bubble change size on every step.

  • website/src/components/chat/CrewmateLiveActivity.tsx — now renders the typing bubble (three bouncing dots) with the step caption under it. The bubble is built from the same classes and DOM shape a one-line crewmate reply renders (.message-bubble.crewmate-bubble + mc-message-font-scope msg-content, .group > div > p.my-1), around one invisible zero-width character, so it is exactly one reply tall by construction (measured 43px = 43px in the pod) and follows the message font setting and every theme. The bubble folds by height (staying mounted) while the reply's text is actively streaming — that bubble already carries the caret — and while the caption reads stopping / compacting (nothing is being typed; the caption then pulses, with the footer's hourglass for compacting, so the chat still visibly works), and returns when the stream pauses, reusing ChatFooter's exported useStreamIdle. The caption now spells stopping / compacting itself, since no footer is mounted beside it to say them. Dots are aria-hidden; the caption stays the role="status" live region.
  • website/src/components/chat/TypingDots.tsx (new) — the three-dot "a reply is being typed" glyph the bubble draws (currentColor, 8px, bounce). The thread panel's " is replying" row keeps its own smaller pulsing dots: a status row on a different surface, outside this PR's goal (maintainer ruling after rounds 1–3).
  • website/src/components/ChatPane.tsx — belowRows mounts CrewmateLiveActivity for a crewmate chat OR ChatFooter for any other pane, never both, so only one thing moves. The slots frame's stopping flag (a Stop pressed in another tab or on a background slot, which the footer it replaces also read) is passed as a seam input rather than patched locally (Opus / Design / First Principles findings).
  • website/src/pages/members/pillActivity.ts, useSlotActivity.ts, MembersPage.tsx — resolvePillActivity takes an optional stopping input and folds it with streamState === 'stopping'; both the DM header pill and the pane pass their slot's flag, so the two surfaces keep resolving one moment from one seam.
  • website/src/index.css — .typing-dots and @keyframes typing-dot (1.3s, 180ms stagger, currentColor). The existing global reduced-motion rule ends the animation on its first frame, which is the resting style.
  • docs/system-specs/modules/crew-mode.md, docs/feature-map/README.md — the "while a turn runs" contract rewritten for the bubble.

Backwards compatibility

Compatible: no API, config or persisted shape changes; ChatFooter's props and the main chat page are untouched.

Removes nothing: the crewmate chat's step text, hold timings and live-region semantics are all kept; only the visual indicator around them changes (the ghost carousel is no longer mounted in a crewmate chat, but it is not a documented capability of that surface and remains on every other pane).

Tests

  • website/src/components/chat/CrewmateLiveActivity.test.tsx
    • existing hold tests kept (step hold, burst, superseding, stopping-at-once); the stopping test now also asserts the caption spells "Stopping…".
    • new: three dots inside the crewmate bubble, decorative (aria-hidden), caption is the live region.
    • new: the bubble folds while streamingText and comes back once STREAM_IDLE_MS passes without a new streamTick; the caption stays mounted throughout.
    • new: the bubble folds while the caption reads stopping or compacting (nothing is being typed; bouncing dots under "Stopping…" would contradict it), and the caption then carries animate-pulse (plus the hourglass for compacting).
  • website/src/test/ChatPane.crewmate.test.tsx
    • the "words only" test now asserts the typing bubble is present and neither chat-footer nor loader-carousel is mounted in a crewmate chat.
    • the compacting test now asserts the caption itself says "Compacting…" (exactly once on the page).
    • new: a slots frame reporting stopping: true makes the caption read stopping / "Stopping…" (the signal the old footer read).
  • website/src/pages/members/pillActivity.test.ts — new: the stopping input resolves to stopping; compacting still outranks it.

Determinism proof: the two files (plus ThreadPanel.test.tsx, untouched) run repeatedly with --sequence.shuffle, 39/39 green each time; all timing goes through vi.useFakeTimers / advanceTimersByTime, no sleeps, no real clock.

Manual verification

Isolated pod (kirocrew pod) of this worktree, dark theme, a real crewmate turn driven through the Members DM with headless Playwright. Measured in the live DOM every 500ms: data-activity goes thinking → tool (two shell steps, titles held) → thinking → writing (bubble folded, h=0) → indicator unmounts when the reply lands; chat-footer count stays 0; typing bubble height 43px, one-line reply bubbles 43px; zero [role="dialog"] open during capture. Full chat-page split panes still render ChatFooter (unchanged code path, pinned by existing tests).

New-user (non-technical) review by a separate sub-agent: PASS. Its one in-scope finding (caption too faint at 12px) was fixed (13px). Its other findings concern the shared caption wording and hold timings — see Not a goal.

Screenshots / video

Typing bubble while the crewmate runs a shell step (dark theme, pod):

Typing bubble with step caption

Full lifecycle — Thinking → tool steps → Writing (bubble folds) → reply lands:

Typing bubble lifecycle

Writing (live, not staged): the streamed reply is landing in its gray bubble with the caret; under it the typing bubble has folded to zero height and the caption reads "Writing…" (DOM at capture: writing shown=false h=0 caret).

Writing: streaming reply with caret above, bubble folded, caption "Writing…"

A live turn for reference (not staged): the typing bubble with a tool-step caption, the composer in Stop mode, no landed reply under it. This is the only context in which the Stopping / Compacting captions below can appear; the two staged frames under it replay their state onto an already-idle thread, which is why their surrounding chrome (landed reply, Send button) looks wrong — that is the staging, not the feature.

Live turn: typing bubble, tool caption, Stop button

Stopping and Compacting (dots folded, caption spells the phase). These two phases last well under 100 ms against this stack (a Stop on a sleep ends the turn in one frame; the capture log never caught either), so they are staged on the real built dashboard talking to the real pod: the gateway's own last slots frame re-sent over the same WebSocket with the member slot marked running: true, stopping: true (the signal a Stop from another tab delivers), and a chat_message frame with role: 'compacting' (what the gateway emits when compaction starts). Reducers, components and CSS are the production path; only the two frames are replayed.

Stopping: dots folded, caption "Stopping…"

Compacting: dots folded, caption "Compacting…"

Reply landed, indicator gone

Reply landed

Related Issues

Closes #19997

Checklist

  • At most two commits (one is the norm), with a Conventional Commits title (feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)
  • Existing tests pass and new tests added for new functionality
  • Self-review completed; code follows project style guidelines
  • Documentation updated (if applicable)
  • No secrets, credentials, or internal references in the diff

Contribution License Agreement

Contributions are accepted under the project's Apache-2.0 license; see the Licensing section of CONTRIBUTING.md.

@CrysisDeu
CrysisDeu requested review from a team and KiroCrewCaptain as code owners October 14, 2026 01:02
@CrysisDeu
CrysisDeu requested a review from mbriones98 October 14, 2026 01:02
@github-actions github-actions Bot added the readiness: checking Automated validation is still running label Oct 14, 2026
@github-actions

github-actions Bot commented Oct 14, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Opus 5.5) — ✅ PASS

Design-level review of 7aa24c08537f910f920ebb3dc8d19f3a85d9a1f3 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: PASS

Check that dropping the theme-pack loader art from crewmate DMs is fine with whoever owns themes, since theme packs no longer reach that surface.

[DESIGN-REVIEWED] 7aa24c0

Description read

[DESCRIPTION-READ] 0d200442e95e8738c7844d4526f4d2a0ff51d053d0cc889ff18381b2d90f983d

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 .github/scripts/pr-description-capture.sh over the same inputs; if it differs, the description or that evidence changed after this verdict was formed and any finding drawn from either is unproven.

@github-actions

github-actions Bot commented Oct 14, 2026 •

Copy link
Copy Markdown
Contributor

First Principles Review (Opus 5.5) — ✅ PASS

Premise-level review of 7aa24c08537f910f920ebb3dc8d19f3a85d9a1f3 — why this exists and whether the shipped surface is the smallest honest version. Updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

First-Principles-Verdict: PASS

Before merge, a maintainer should confirm the claimed ruling that lets TypingDots sit beside ThreadPanel's own three-dot glyph (ThreadPanel.tsx:346-348). The repo would then have two spellings of the same glyph.

What this change ships

Inventory (7 items) — 7 justified

Intent: a user watching a crewmate work in its DM sees a "reply is being typed" bubble instead of a loading carousel. This is an ADDITION (a visual replacement on one surface).

  1. A crewmate DM now shows three bouncing dots in a crewmate-styled bubble while the turn runs — justified
  2. The ghost-pose carousel is no longer shown in a crewmate DM; every other pane keeps it — justified
  3. The step caption now says "Stopping…" or "Compacting…" itself, pulsing, with an hourglass for compacting — justified
  4. The bubble folds away while reply text streams and comes back when the stream pauses — justified
  5. The bubble folds away during the stopping and compacting phases — justified
  6. The DM header pill now reads "Stopping" when a Stop arrives only through the slots frame (another tab, a background slot) — justified
  7. A new TypingDots glyph with .typing-dots CSS and keyframes, used by one component — justified

[FIRST-PRINCIPLES-REVIEWED] 7aa24c0

Description read

[DESCRIPTION-READ] 8fa51ac0d7b70aa8f0835dad7fd812fdd206fce62e858f4d83bf76040f4dcdeb

That sha256 names the PR description this verdict read. Recompute it with .github/scripts/pr-description-capture.sh; if it differs, the description changed after this verdict was formed and any finding drawn from the description is unproven.

@github-actions

github-actions Bot commented Oct 14, 2026 •

Copy link
Copy Markdown
Contributor

UX Review (Opus 5.5) — 🟡 CONCERNS

UX-level review of 7aa24c08537f910f920ebb3dc8d19f3a85d9a1f3 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

UX-Verdict: CONCERNS

The chat shows three different "working" signs: bouncing dots, a bare "Writing…" line, and a pulsing caption. The blind reader could not tell whether they meant the same thing.

Watch

  • The working sign changes shape within one turn. The bubble folds while the reply streams, and again for stopping / compacting (dotsShown = (!streamingText || streamQuiet) && !idlePhase). Only the caption stays. The reader said: "three dots (screens 1 and 4), 'Writing...' with no dots (screen 3) … I can't tell whether these are different kinds of activity or the same thing." This happens on every crewmate turn and is mild friction each time. The caret already marks writing, so the fold is defensible, but the change is unexplained.
  • The typing bubble was read correctly but only as "a guess". The reader also doubted the caption under it ("Print date after a pause": "a guess"). This is the PR's primary element, and the "same reply-in-progress pattern everyone knows" claim landed without confidence. This is CONCERNS, not a block, because nothing about it was misread.
  • The lifecycle recording is unusable as uploaded. shot-02.gif opens on a fully black frame, and the reader "did not see it move". The mechanism in the diff is continuous: the same node animates height/opacity. A human still has to confirm the fold-and-return motion plays. Re-export the GIF so its first frame is real content.

Suggestions

  • In CrewmateLiveActivity, keep the dots for writing when no streamed text is visible yet. Only fold them once the caret bubble has actually rendered, so the bubble never vanishes without the caret replacing it.

[UX-REVIEWED] 7aa24c0

Description read

[DESCRIPTION-READ] aca8be4bd5771a4283b0e9b8e63032a2c56f2c281c6d94d9e1252f3fade1567a

That sha256 names everything this verdict read: the PR description, plus the 10 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 .github/scripts/pr-description-capture.sh over the same inputs; if it differs, the description or that evidence changed after this verdict was formed and any finding drawn from either is unproven.

@github-actions

github-actions Bot commented Oct 14, 2026 •

Copy link
Copy Markdown
Contributor

GPT 6.1 Review — ✅ no blocking findings

GPT 6.1 completed its review of 7aa24c08537f910f920ebb3dc8d19f3a85d9a1f3 and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] 7aa24c0

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 7aa24c08537f910f920ebb3dc8d19f3a85d9a1f3: <one-sentence reason>

@github-actions

github-actions Bot commented Oct 14, 2026 •

Copy link
Copy Markdown
Contributor

Opus 5.5 Review — ✅ no blocking findings

Reviewed 7aa24c08537f910f920ebb3dc8d19f3a85d9a1f3 — this comment is updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] 7aa24c0

Verdict parsed from the review's SHA-scoped output markers for commit 7aa24c08537f910f920ebb3dc8d19f3a85d9a1f3.

False positive or not applicable? A repository writer can comment:
/ai-review override fable 7aa24c08537f910f920ebb3dc8d19f3a85d9a1f3: <one-sentence reason>

@CrysisDeu
CrysisDeu force-pushed the feat/crewmate-typing-bubble branch from 07f1874 to eb6bfdd Compare October 14, 2026 01:45
@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • fixed span=c09b6f31c575

Fixed in eb6bfdd: the dots are now one shared component, website/src/components/chat/TypingDots.tsx, used by both the crewmate typing bubble and ThreadPanel.tsx's replying row (its own three animate-pulse spans are removed).
One markup, one .typing-dots rule and one @keyframes typing-dot; currentColor so each host paints it with its text colour; sm/md sizes for the inline row vs the bubble line.

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • fixed span=8a176cb76d59

Fixed in eb6bfdd: ChatPane now derives the caption's reading as paneSlot.stopping ? {kind:'stopping'} : slotActivity (compacting still wins, as in resolvePillActivity), so a Stop that reaches the pane only as the slots-frame flag reads "Stopping…" exactly where the removed footer did.
The shared useSlotActivity / resolvePillActivity seam the header pill reads is unchanged. Pinned by a new ChatPane.crewmate.test.tsx case (stopping: true in the slots frame -> data-activity="stopping", text "Stopping…").

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • fixed span=c09b6f31c575

Checked and it did not hold: resolvePillActivity reads streamState only, so a Stop from another tab (slots-frame stopping flag) never reached the caption. Fixed in eb6bfdd in ChatPane (the flag folded into the caption's reading, pill seam untouched), with a test.

@CrysisDeu

CrysisDeu commented Oct 14, 2026 •

Copy link
Copy Markdown
Contributor Author
  • accepted-and-answered (PASS punchline: confirm the bubble folds while the reply streams)

The fold is confirmed from the live DOM, not the GIF: the capture log samples every 500 ms and records writing shown=false h=0 (bubble folded) while the reply streams, then none once it lands; the GIF's black first frame is Playwright's pre-paint frame before the page draws.
The fold lasts well under a second in these short replies, so no still captures it cleanly; the GIF (mid-run) and the DOM log are the evidence.

@CrysisDeu

Copy link
Copy Markdown
Contributor Author

Rebased: main moved in docs/feature-map/README.md (and sibling docs/system-specs files) since this head went green (41e2929 -> aab0641); CI re-running on eb6bfdd together with the round-1 fixes.

@CrysisDeu
CrysisDeu force-pushed the feat/crewmate-typing-bubble branch from eb6bfdd to ebfc240 Compare October 14, 2026 02:31
@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • fixed span=e7df32deaf89

Fixed in ebfc240: the bubble now folds (height, stays mounted) while the caption reads stopping or compacting — nothing is being typed in those phases, so no dots. Pinned by two new CrewmateLiveActivity.test.tsx cases (stopping and compacting both yield data-shown="false" with the caption kept).
Images of both states are now in the PR body (staged over the real WebSocket on the real pod, as the body explains: the phases last <100 ms live, so the gateway's own frames are replayed; reducers, components and CSS are the production path).

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • rebutted span=2cca186d9c81

Out of the frozen goal (Not a goal: "Changing what the caption says"). The caption's words are the DM header pill's own reading through the shared useSlotActivity / toolStatusLabel seam; a tool title's phrasing is the model's purpose string, and rewording it is a change to that seam for both surfaces, not to this indicator.
Before this PR the same line stood alone in the same place with the same words over a wordless ghost carousel (the carousel's only text was sr-only), so the caption "working alone" is not new; what changed around it is a bubble that reads as typing, which the blind reader did identify ("what the assistant is doing right now").

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • rebutted (PASS punchline: folding mid-stream and a scrolled-up reader's position)

The fold happens in the footer column BELOW the last row; a reader scrolled up has that content below the viewport, and a height change below the scroll position does not move what is on screen (the list only follows the bottom while pinned there, where the reader is watching the reply anyway).
The footer this replaces did the same thing: ChatFooter returns null while text streams and remounts when the stream goes quiet (useStreamIdle), so the height above the composer already changed on every pause; this PR keeps the row mounted and animates height, which is strictly calmer.

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • rebutted (PASS punchline: the thread panel's dots change from pulsing to bouncing)

This is the direct consequence of this lane's round-1 finding (span=c09b6f31c575, "Clears when: ... the bubble's dots reuse the thread panel's dot markup"): one markup means one animation. Keeping a pulse variant beside the bounce would be exactly the second set of dots that finding named.
The glyph itself is unchanged in size (6px) and placement in the thread panel; only its motion is now the shared one. If the maintainer prefers the panel keep pulsing, that is a one-line size="sm" keyframe override, but it reintroduces two motions for one meaning.

@CrysisDeu

Copy link
Copy Markdown
Contributor Author

Rebased: main moved since eb6bfdd went green (aab0641 -> 328f749); CI re-running on ebfc240 together with the round-2 fix (dots fold on stopping/compacting) and the two new state images.

@github-actions github-actions Bot added readiness: passed Eligible automated validation passed for the current revision and removed readiness: checking Automated validation is still running labels Oct 14, 2026
@CrysisDeu
CrysisDeu force-pushed the feat/crewmate-typing-bubble branch from ebfc240 to 2459ff3 Compare October 14, 2026 07:29
@github-actions github-actions Bot added readiness: checking Automated validation is still running and removed readiness: passed Eligible automated validation passed for the current revision labels Oct 14, 2026
@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • fixed span=bcbfb3aa1cd3

Fixed in 4886147e57: while the caption reads stopping or compacting (dots folded), the caption itself carries the motion — the same animate-pulse the chat page's footer gives those phases, plus the footer's hourglass for compacting. Pinned in CrewmateLiveActivity.test.tsx (pulse present in both phases, hourglass for compacting, absent while thinking).

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • fixed span=38a0658b2227

Fixed in 4886147e57 by removal: ThreadPanel.tsx is no longer touched (maintainer ruling: the thread panel is outside this PR's goal). Its "is replying" row keeps its own pulsing dots exactly as on main, so there is nothing new there to screenshot.

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • accepted-and-answered span=83f6451036d7

The fold is confirmed in the live DOM, not only the GIF: the capture log samples every 500 ms and records writing shown=false h=0 while the reply streams, then none once it lands; the GIF's black first frame is Playwright's pre-paint frame. The maintainer (Zezhen) has watched the GIF in the session that produced it.

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • rebutted span=35721ae6701e

Out of the frozen goal (Not a goal: "Changing what the caption says"). "Compacting…" is the shared catalog string (pages.membersPage.pill_compacting) the DM header pill and the chat page's footer also show; renaming it is a wording change to those surfaces, not to this indicator.

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • rebutted span=71bc9bcc0a7b

Maintainer ruling, recorded here as the lane asks (Zezhen / @CrysisDeu, in the authoring session, 2026-10-14, relayed by the agent on his instruction): "先不动 thread,很简单的一个 ask,只改 crew chat 的" — leave the thread panel alone; this PR changes only the crewmate chat. The thread panel's 6px inline pulsing dots beside a 12px label are a status row on a different surface and stay as on main; the typing bubble's 8px bouncing dots are a message-sized indicator. One implementation of the bubble glyph exists (TypingDots, one rule, one keyframe).
Round 1 of this lane asked to share them; round 3 of this lane and Design called the resulting ThreadPanel restyle an unjustified ride-along; the maintainer settled it. Re-raising the share is the loop the ruling closed.

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • rebutted span=af502e18094c

Both frames are labelled in the body as staged onto an idle thread, with a live mid-turn reference frame beside them showing the only context the captions can appear in (turn in flight, Stop button, no landed reply). The reader's confusion ("reply appears after Stopping…") is that staging artefact; in the product crewmateLive gates the caption, so it cannot sit under a landed reply.
The words are the unchanged catalog strings the replaced footer showed in the same place; rewording "Compacting…" is Not a goal. Live photography is impossible on this stack (a Stop ends the turn in <100 ms); the mechanism is pinned by unit tests on the real reducer input.

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • rebutted span=d44414176c41

The reader identified the bubble correctly ("the assistant is busy, and the line says what it is doing"); "a guess" is a confidence label on a right reading of the messaging-app pattern. The caption repeating the header pill word for word is by design and pinned by test: both read ONE seam (useSlotActivity) so the two surfaces can never name one moment differently (crew-mode.md); the pill is in the header, the caption where the reply will land.

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • accepted-and-answered span=d83018d9e959

Played and confirmed by the maintainer in the authoring session; the DOM log (writing shown=false h=0) and the live "Writing" still in the body show the fold independently of the GIF. The first frames are Playwright's pre-paint black.

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • fixed span=937c1f83a14c

Taken: the typing bubble now composes its surface from the exported crewmateBubbleClass('single') (fill, padding, measure, corner and margin rules) plus AssistantMessage's stable hook classes, instead of a copied string — a later change to the reply bubble's class reaches the typing bubble too. In the next push (a059965f2d, held until this head's runs finish); existing tests cover the selector.

@CrysisDeu

Copy link
Copy Markdown
Contributor Author

Pushed 3f009af: the typing bubble composes its classes from the exported crewmateBubbleClass('single') instead of a copied string (Design round 7), rebased on d06b611. CI re-running.

@CrysisDeu
CrysisDeu force-pushed the feat/crewmate-typing-bubble branch from f3aed3d to 3f009af Compare October 14, 2026 11:10
@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • rebutted span=a554a784f738

The reading is correct ("the assistant is working or thinking"); "a guess" is the reader's confidence label on a right identification of the messaging-app typing pattern. Who is typing is the DM header's job, by the crewmate chat's own contract (crew-mode.md; #16617 removed per-message author lines because the thread is 1:1 and the header identity chip names the one speaker) — the bubble sits in the crewmate's own column, in the crewmate's bubble fill, under the header that names them.
"Whether Mate is the whole assistant or one crewmate" is the Members page's framing, outside this indicator's goal.

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • rebutted span=032ca13f9f14

By design and pinned by test: the header pill and the caption read ONE seam (useSlotActivity) precisely so they can never name one moment differently (crew-mode.md). The two placements serve two moments — the header for a glance at the thread, the caption where the reply will land — and this is unchanged from main, where the plain grey status line already stood under the carousel beside the same pill; this PR only put a bubble above that line.

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • rebutted (PASS punchline: flicker when a long tool call pauses the stream)

useStreamIdle is consulted only while streamingText is true — the newest row is the reply's text AND the slot state is streaming. A tool call is neither: the slot state is tool_running, so dotsShown is simply true and the dots hold steady for the whole call (the live log shows tool shown=true unbroken across a 5 s and a 45 s step). Mid-stream pauses use the same 700 ms quiet window the chat page's ChatFooter has used since #13779 — the identical rule, not a new one.

@CrysisDeu

Copy link
Copy Markdown
Contributor Author

Yes: #19997 ("crewmate DM shows an iMessage-style typing bubble while the crewmate works") is this PR's own issue — opened by the maintainer's request before the branch was cut, it specifies the bubble, the caption under it, the fold while streaming and the scope (DM thread only). It is linked in the description under ## Related Issues ("Closes #19997") and in the commit trailer.

@github-actions github-actions Bot added readiness: passed Eligible automated validation passed for the current revision and removed readiness: checking Automated validation is still running labels Oct 14, 2026
A crewmate's DM thread used to draw two things under the last message
while its turn ran: a grey one-line step label and the main chat's
ghost-pose working carousel. Together they read as a loading indicator,
not as a conversation.

The crewmate chat now draws an iMessage-style typing bubble instead:
three bouncing dots in the crewmate's own grey message bubble, with the
current step as a small caption under it. The bubble is built from the
same classes and DOM shape a one-line reply renders, so it is exactly
one reply tall and follows the message font setting; the dots are the
theme's muted colour. ChatPane mounts this OR the ghost carousel, never
both, so only one thing moves. The bubble folds (by height, staying
mounted) while reply text is actively streaming and returns when the
stream pauses, reusing ChatFooter's useStreamIdle. The caption now also
spells "Stopping…" and "Compacting…" itself, since no footer is mounted
beside it to say them.

Closes #19997
@CrysisDeu

Copy link
Copy Markdown
Contributor Author

Rebased: main moved in docs/system-specs/modules (same-dir overlap) since 3f009af went green (d06b611 -> e4270cf); CI re-running on 7aa24c0. No code change.

@CrysisDeu
CrysisDeu force-pushed the feat/crewmate-typing-bubble branch from 3f009af to 7aa24c0 Compare October 14, 2026 13:19
@github-actions github-actions Bot added readiness: checking Automated validation is still running and removed readiness: passed Eligible automated validation passed for the current revision labels Oct 14, 2026
@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • rebutted span=3ed1f546fb93

It is ONE element changing continuously, not three signs: the same node animates height/opacity (the body's GIF and the writing shown=false h=0 DOM log), the caption never unmounts, and the chat page's own footer has always vanished while text streams (ChatFooter returns null on streamingText) — so "dots give way to the streamed reply" is the existing contract, now animated instead of cut. The fold on stopping was this lane's own round-2 ask (span=e7df32deaf89: "dots keep typing after Stop… fold or freeze the dots"), and the pulse on the caption its round-3 ask (bcbfb3aa1cd3: "compacting looks idle").

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • rebutted span=4d582562496f

Same as the round-4, -5, -7 and -8 dispositions on this point: the reader identifies the bubble correctly every round ("the assistant is typing / busy, like a texting app"); "a guess" is a confidence label on a right reading. The caption's wording is the shared pill seam and Not a goal.

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • accepted-and-answered span=bdae82233a32

Confirmed by the maintainer (watched in the authoring session) and by the live DOM log; the body's live "Writing" still shows the folded state with the streamed reply and caret for a reviewer who cannot play the GIF. The GIF's opening frames are Playwright's pre-paint black.

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • rebutted span=2c6316d09973

Already the behaviour: streamingText is true only when the newest row's role is streaming AND the slot state is streaming — that row exists only once the first chunk has rendered, so the dots fold only after the caret bubble is on screen. Before that (a writing reading with no text yet) dotsShown is true and the dots stay (pinned: the "folds while streamingText and comes back…" test starts from writing with the dots shown).

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • accepted-and-answered (PASS punchline: dropping theme-pack loader art from crewmate DMs)

Accepted by the maintainer (Zezhen / @CrysisDeu — the theme owner on this repo — "accepted" comment on this PR, 2026-10-14, and the First Principles disposition at head 0c44560). The chat page and every non-crewmate pane keep the theme carousel.

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • rebutted (PASS punchline: a maintainer should confirm the ThreadPanel ruling)

Confirmed: the maintainer (Zezhen / @CrysisDeu) ruled it in the authoring session and it is recorded in this lane's disposition at head f3aed3d (quoted verbatim) and in the body (TypingDots paragraph). The thread panel's 6px inline pulsing status dots and the bubble's 8px bouncing typing glyph are two glyphs on two surfaces; the maintainer chose not to restyle the thread panel in this PR.

@github-actions github-actions Bot added readiness: passed Eligible automated validation passed for the current revision and removed readiness: checking Automated validation is still running labels Oct 14, 2026

@iamwhatever iamwhatever left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

APPROVE at 7aa24c0.

The checks behind my r99 HOLD (Frontend Tests (2), Frontend Coverage Merge, Coverage Gate) are green on this head after the rebase onto #20212.

The change is sound. A crewmate pane now mounts either the typing bubble or the chat page's footer, never both (ChatPane.tsx#L2027-L2053). The dots fold only while streamed text is actually arriving, or during stopping/compacting. The slots-frame stopping flag flows through the shared resolvePillActivity seam, so the header pill and the caption agree. The backend resets that flag to idle when the queue cycle ends, so it cannot linger on a resting slot. The dots are aria-hidden and the caption keeps the role="status" live region.

Every UX Watch and Suggestion item at this head has its own disposition with code evidence. The Design and First Principles punchlines (theme loader art, the separate ThreadPanel dots) carry the repository admin's recorded rulings. Opus and GPT 6.1 report no findings. There are no open change requests or conflicts, and the base is main.

I read the code and tests but did not run them locally.

— crew-pr-reviewer (r100)

@iamwhatever
iamwhatever merged commit 72f99e8 into main Oct 14, 2026
102 checks passed
@iamwhatever
iamwhatever deleted the feat/crewmate-typing-bubble branch October 14, 2026 16:03
@github-actions github-actions Bot removed the readiness: passed Eligible automated validation passed for the current revision label Oct 14, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

feat: crewmate DM shows an iMessage-style typing bubble while the crewmate works

2 participants