Repository navigation
Conversation
In the main chat the user's messages and the agent's replies look almost the same: a neutral bg-card bubble beside a reply with no bubble. A crewmate's chat already fills the user's bubble with the theme accent through tone="accent" on UserMessage. Pass that tone at every seam the main chat draws the user's row from: ChatPage's bubble row, the SDK's default user row (split panes, Side Chat, app chat embeds) and the crew window's row. The pinned-prompt card is a pixel copy of the bubble, so it takes the same hooks and the hand-off stays invisible. A one-pager records the decision, the crew-mode spec says where the accent applies, and before/after captures sit under temp-screenshots/.
Design Review (Opus 5.5, fork) — ✅ PASSDesign-level review of Design-Verdict: PASS Nothing blocks: one existing tone is passed at each place the main chat draws a user bubble, the screenshots are committed, and a revert undoes it cleanly. Notes
[DESIGN-REVIEWED] 2e9c7b0 Description read[DESCRIPTION-READ] 30c17f133a658323b6195cde260da7f20eac98e3b4d29d36cd1b56346fd9e84d 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 |
UX Review (Opus 5.5, fork) — ✅ PASSUX-level review of UX-Verdict: PASS Nothing blocks. The screenshots show the accent bubble in the main chat, the pinned card, Edit & Resend, and a phone in both themes. No first-time reader has looked at them, because this lane reviews forks without a blind read. Notes
[UX-REVIEWED] 2e9c7b0 Description read[DESCRIPTION-READ] 09081c46326fe0961f970d0419eb5eeb4c6c1306831d4de61249de76ae3f9727 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 |
First Principles Review (Opus 5.5, fork) — 🔴 BLOCKPremise-level review of First-Principles-Verdict: BLOCK product-shape change without accepted RFC: every user's main chat, and every app's default chat embed, switches from a neutral user bubble to an accent-filled one. The only decision record is the RFC this same diff adds. What this change shipsInventory (8 items) — 5 justifiedIntent: make your own messages in the main dashboard chat easy to tell apart from the agent's replies. ADDITION.
Blockersproduct-shape change without accepted RFC: the main chat's user bubble becomes accent by default (items 1, 3) [FIRST-PRINCIPLES-REVIEWED] 2e9c7b0 Description read[DESCRIPTION-READ] 43042b40968fb19daa0731b04f2425363d527dc66f68e19fa912bf57e3dbdc8d That sha256 names the PR description this verdict read. Recompute it with |
GPT 6.1 Review (fork) — ✅ no blocking findingsReviewed Review detailsNo findings. |
Opus 5.5 Review (fork) — ✅ no blocking findingsReviewed |
bolichen97
left a comment
There was a problem hiding this comment.
Needs a design decision before merge. The code itself looks good.
Head reviewed: 2e9c7b0. Lanes read on this head: Design PASS, UX PASS, Opus no blocking findings, GPT no blocking findings, First Principles BLOCK. Readiness is action required, and CI is still running.
What I checked:
- The four
tone="accent"seams (ChatPage row, SDK defaultuserrow, crew window row, pinned prompt card). Each one reuses the existinguser-bubble-accenthook unchanged. Themsg-contentclass added to the pinned card is what the existing selectors inchat-transcript.cssneed, and the CLI guard still turns it off. - A steer with its badge still wins over the tone in
UserMessage, so the steer tint is unchanged. - Each seam has a test that goes red if its tone is dropped.
Why this is a comment and not an approval: the linked issue #23358 is tier:T3 and is not on the current plan. A T3 change needs a design sign-off from a maintainer. The First Principles lane blocks on the same point: the only decision record is the RFC this PR adds, and the RFC index says an RFC shipped beside its change proposes a decision but does not record one. One way forward is to land the RFC on its own as draft and let a maintainer accept it, then rebase this PR. A maintainer can also record the decision on this PR directly.
🔴 0 must-fix · 🟡 1 should-fix (inline). Not repeated: the product-shape blocker the First Principles lane already states.
| @@ -0,0 +1,77 @@ | |||
| --- | |||
| title: The main chat's user bubble uses the accent tone | |||
| status: implemented | |||
There was a problem hiding this comment.
🟡 The RFC records itself as implemented before anyone accepted it or anything merged
The front matter sets status: implemented, but doc-pr is empty, implementation-prs is [], and the index row says what is "actually on main" while the change is still open. The status vocabulary defines implemented as "every phase is verifiably on main". A reader of the index sees an accepted and shipped decision that does not exist yet. If this PR waits or closes, the index stays wrong.
status: implemented
...
doc-pr:
implementation-prs: []Why it matters: the index is the record other reviewers and the First Principles lane trust for product-shape decisions. A self-set implemented row makes the record say a decision was taken when it was only proposed.
Suggestion: set status: draft here and in the README row, and describe the row as proposed. After a maintainer accepts the design, move it to accepted, then fill doc-pr and implementation-prs when the change lands. If a maintainer already agreed, name that in the prose status line.
Rule: docs/request-for-change/README.md "Status vocabulary" and "Keeping this honest"
|
Closing this. A theme can already do it with no code change: an |
Problem / Motivation
Goal: In the main dashboard chat, the user's own messages are filled with the theme accent, so they stand apart from the agent's replies at a glance.
In the main chat your message is a neutral gray card and the agent's reply has no bubble at all. In a long session the two look almost the same, and a short question gets lost between long replies (#23358). A crewmate's chat already solved this with an accent-filled user bubble (#17839, shipped in #17918).
Why it matters
The main chat is where most conversations happen. Scrolling back through a long session, you look for your own questions first, and today you have to read each block to find them. Dark and light themes are both affected.
Not a goal
What changed (motivation → approach → change)
Your messages in the main chat are now filled with the theme accent, so you can tell them from the agent's replies at a glance. It is the same bubble a crewmate's chat already draws.
Four files draw your message, and each one now asks for the accent tone. That reaches these surfaces:
pages/chat/page/transcriptRendering.tsx)app-sdk/messageRenderers.tsx, the shareduserrow)pages/chat/crew-window/crewWindowRenderers.tsx)pages/chat/PinnedPrompt.tsx)The docs say where the accent applies:
docs/system-specs/modules/crew-mode.md, plus the comments onUserMessage'stoneand on the CSS hook instyles/chat-transcript.css. A one-pager records the decision, astier:T3asks:docs/request-for-change/rfc-main-chat-accent-user-bubble.md, indexed in the directory README.Backwards compatibility
Compatible:
tonealready exists and stays optional, and only class names on four elements change. No prop, key, type or row kind is renamed or removed. An app that wants a neutral bubble can still register its ownuserrow.Removes nothing: the neutral
defaulttone stays the component default, and every reader of.user-bubble,.message-bubbleanddata-testid="pinned-prompt"still finds them.Tests
website/src/pages/ChatPage.transcriptRendering.test.tsx: the main chat page's user row drawsuser-bubble-accent. Fails if the page stops passing the tone.website/src/test/UserMessage.accentTone.test.tsx: the user row that a split pane, the Side Chat and a crew window each resolve (each registry built the way its host builds it) draws the accent. Fails if the shared row or the crew window row drops it. The file's comments now describe the main chat as well.website/src/test/kiroLightShellHooks.test.ts: the pinned card's class pin names the accent hooks, so the card cannot drift from the bubble.crewmateBubbles.test.tsandChatMessageList.test.tsx, and a shuffled order (--sequence.shuffle --sequence.seed=2746).Proof: not provable -- frontend test (prove.py runs Python only)
Manual verification
The real ChatPage in Chromium, with no gateway: a throwaway capture page mounts the page on a fixture transcript and Playwright answers its API calls. Each check compares a computed colour with the theme's own token, in kiro-dark and kiro-light:
var(--accent), textvar(--accent-fg)var(--accent-fg); code keeps its own patchvar(--accent-fg), excerpt a 90%--accent-fgmixvar(--accent-fg)var(--accent-subtle)var(--accent)var(--accent)data-ui="cli"), both themesmainNearby inputs that reach the same code:
renderUserContentpath, inside the same token scopeScreenshots / video
The real ChatPage, the same fixture on both trees.
main)Phone, 390×844, dark, after:
Pinned prompt and Edit & Resend (dark, after)
Related Issues
Closes #23358
The accent tone this reuses: #17839, shipped in #17918.
Sibling sweep
N/A -- a feature, not a fix.
Pattern harvest
N/A -- a feature, not a fix.
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)Contribution License Agreement
Contributed under the project's Apache-2.0 license.