Repository navigation
cmux NEXT composer: remove the bottom-left harness label - #24938
Merged
teamleaderleo merged 2 commits intoOct 25, 2026
Merged
Conversation
Contributor
|
cmux status (one comment, edited in place; each line is updated by the job that owns it) Review App of this head (app.zip, kept 1 day)
Unclear: CI failed on
Failures:
Matched log linesNot re-run automatically: Written by red: cmux-next checks (god files, concurrency, crash safety, l10n); no finished base run on cb1135a to compare; do not merge until green; see lane-rules Tour of Gallery before/after for
States
Pick other states with a |
|
Found 14 test failures on Blacksmith runners: Failures
...and 4 more test failures. View all on Blacksmith: run 40148239893, run 40148239986
|
![Fix with [code]smith](https://pr-comments-assets.blacksmith.sh/codesmith/fix-with-codesmith-light-3x.png)
teamleaderleo
commented
Oct 24, 2026
teamleaderleo
left a comment
teamleaderleo
left a comment
Collaborator
Author
There was a problem hiding this comment.
Independent read-only review — SHIP
Exact head: 5ea61848849d413defee3025aa41d46fca73efeb against origin/feat-cmux-next (a4d0757b114c0b1d8aa6ee2e0b44527728d7e658).
No P1/P2/P3 user-visible defect found. The diff deletes ComposerHarness and removes only its ComposerPickers wiring, footer CSS, and fit step; ModelPicker remains the harness rail and its onHarness/combo paths are unchanged. Machine selector/layout paths are untouched.
Evidence: targeted row/footer unit tests 10 pass; webview typecheck passes (2622 files); git diff --check passes. Browser suites were attempted with CMUX_BROWSER_TESTS=1 but local Chromium/WebKit were unavailable, so those tests were skipped. Full check/lint were not clean due unrelated pre-existing generated/system warnings outside this diff.
Collaborator
Author
|
Bead |
teamleaderleo
force-pushed
the
codex/nx-composer-harness-20261024
branch
from
October 24, 2026 20:48
5ea6184 to
319323d
Compare
teamleaderleo
commented
Oct 24, 2026
teamleaderleo
left a comment
teamleaderleo
left a comment
Collaborator
Author
There was a problem hiding this comment.
Independent read-only review: SHIP
Exact head: 319323d7b57603c254099b4aa6b9bd3a59f4bdaf
Base: cb1135a646229cf614e8e1b403ff4b88183c4c45
The cumulative diff removes only the composer’s bottom-left harness label while preserving harness selection in the model-picker rail and leaving machine selection unchanged. Focused unit tests (10), typecheck (2,624 files), and diff check pass. Browser tests registered but were skipped because Chromium/WebKit engines are unavailable on the review host. Full checks stop only on pre-existing base warnings.
Collaborator
Author
|
Review (Codex Astra xhigh, correctness first): SHIP at ea00d55 Independent exact-head review against base cb1135a.
|
This was referenced Oct 25, 2026
Contributor
|
The feat-cmux-next push run https://github.com/manaflow-ai/cmux/actions/runs/40149382418 at a4c4d46 failed: cmux-next generated files. |
Contributor
|
The feat-cmux-next push run https://github.com/manaflow-ai/cmux/actions/runs/40149412279 at 19a5504 failed: cmux-next generated files. |
Contributor
|
Shipped in:
|
teamleaderleo
added a commit
that referenced
this pull request
Oct 25, 2026
Base removed the quiet row's harness label (#24938), so this PR's harness-fit work is dropped: composer-footer-harness.test.ts takes base's version and the 460 px harness container query goes with it. The quiet row's context ring keeps base's legacy 4 px rule; composerFooterShared.css's 2 px margin outranks it and already closes the loose gap. Agent-Session: https://cmux.com/deeplink/agent?harness=claude&session=01a174e4-2849-7aa2-bd71-289c4ec9e04f&host=huge-blue
teamleaderleo
added a commit
that referenced
this pull request
Oct 25, 2026
…ase's row Base took the harness label out of the quiet row (#24938) and gave the context affordance one compact size with a 2 px margin (composerFooterShared.css), so: the row now keeps every whole label down to a 244 px composer (measured on the fleet: panes 280 and 300 read "Opus 5.5 Medium"), the ring's glyph sits 15 px after the effort, and the shared focus ring is asserted from the CSS because the gallery stage loads no pages/shared/desktop.css. composer-app-font also reads base's new composerFooterShared.css. Agent-Session: https://cmux.com/deeplink/agent?harness=claude&session=01a174e4-2849-7aa2-bd71-289c4ec9e04f&host=huge-blue
teamleaderleo
added a commit
that referenced
this pull request
Oct 29, 2026
… own room (#24151) * cmux NEXT composer test: the model button is quiet, its bolt is inside it, and the row fits on its room (failing) Leo 2026-10-21, of #23742's after-shot: "WTF IS THIS UGLY BULLSHIT. WE ALREADY FIXED THE FONTS. THE BOLT DOESN'T GO THERE." And live-qa on nightly 3a63d6d: a 410 px composer read "Opus Med" with about 250 px of the quiet row free. - composer-fast-toggle: no harness draws a speed button in the box; Fast mode is the model menu's last row for Claude and Codex; a running Codex fast tier is a small outline bolt inside the model button, before the name; Claude never draws a bolt. - composer-model-quiet (Chromium, the real composer on the gallery stage): the bolt's place and line; no fill, border, ring or outline at rest; the effort a step back of the name; keyboard focus a pill and a hairline, not the bright ring the capture caught; the context ring's glyph about 8 px after the effort on the buttons' center line; the button, its labels and the mode in the composer's font; "Opus 5.5 Medium" kept at 410, 324 and 264 px, and whole-unit steps with nothing overlapping at 204 and 164 px. - composer-app-font: the composer names the app's UI font once, no composer rule names another family, and the quiet row's buttons inherit it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?harness=claude&session=01a16027-99fc-73f0-9482-5a4babfe1b0a&host=cosmic-gray * cmux NEXT composer: Codex's bolt goes inside a quiet model button; the quiet row fits on its own room Leo 2026-10-21, of #23742's after-shot ("GPT-6 Astra Medium" in a bright outline, a bolt in the box by the mic, a loose ring): "WTF IS THIS UGLY BULLSHIT. WE ALREADY FIXED THE FONTS. THE BOLT DOESN'T GO THERE." Bolt. FastToggle.tsx (#23742's bolt button left of the mic) and its CSS are gone, with the five strings only it used. Fast mode is the model menu's last checkbox row for every harness that offers it, Codex included. While a Codex fast tier runs, the model button draws a 12 px outline bolt before the model's name, in the name's ink, and its tooltip says Fast Mode; Claude Code draws no bolt anywhere. The stale "no bolt in the composer" comments now say no speed button, and the gallery's reasoning-codex shows the menu's Fast mode row, with reasoning-codex-fast for the bolt. Quiet button. The outline in the shot was keyboard focus, not rest: the gallery's play closed the menu with Esc, which hands focus back to the button, and the shared ring (desktop.css, 1.5 px at 48%) drew around bare text. At rest the button already drew no edge. In the quiet row its focus is now the row's soft pill and a 1 px hairline at 22% (Increase Contrast keeps the shared ring), the effort sits one step back of the name in every theme, and the context ring lost its 8 px margin, so its glyph is 9 px after the effort (was 17) on the buttons' center line. Fonts. Nothing regressed: the composer names system-ui (SF Pro in the app's WebKit) since #20374, and #23742 touched no font. The shot came from headless Linux Chromium, which has no SF Pro and fell back to Liberation Sans for the whole page. The quiet row's buttons sat outside `.acpmux-composer-box button { font: inherit }`, so they drew in the engine's control font; they now inherit the composer's family, and composer-app-font holds it. Truncation. COMPOSER_FIT_STEPS shortened on the composer's width (#20802's one row). Since #22285 the quiet row holds only the mode, the model button and the ring, so its width steps were stale: a 410 px composer read "Opus Med". The quiet row now takes no width step (paneWidth is unbounded for .acpmux-composer-foot, and the container queries skip it); useRowFit fits it on the units' natural widths against the row's measured room, so it keeps "Opus 5.5 Medium" down to a 264 px composer here and shortens whole units only when the row runs out. The box's own row keeps its steps. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?harness=claude&session=01a16027-99fc-73f0-9482-5a4babfe1b0a&host=cosmic-gray * cmux NEXT composer: no dot between the model and its effort Leo 2026-10-21: "BRO WTF NO WE FUCKING HATE THE DOT", of Lawrence's build reading "Opus 5.5 • Medium". The one dot the button can draw on base is the agent-update dot (hq#2569 #12, `.acpmux-model-update-dot`): ModelPicker's chip face puts it after the model's name, so with Claude Code or Codex outdated it sits between the name and the effort. No label formatter writes a "•" or "·" into the button; the fit steps draw whole labels. The dot is gone. The update stays in the button's tooltip and in the menu's and picker's Update row. composer-fast-toggle checks, for Claude and Codex, Fast on and off, with and without an update on offer, that the button's text and every fit label carry no "•" or "·" and nothing sits between the name and the effort; composer-model-quiet checks the rendered button reads "Opus 5.5 Medium" at 410, 324 and 264 px and carries no dot at the narrow steps. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?harness=claude&session=01a16027-99fc-73f0-9482-5a4babfe1b0a&host=cosmic-gray * cmux NEXT composer test: the quiet row keeps whole labels at every width; the in-box row keeps its steps composer-row-steps measured whichever row held the model button, and its one table still held the width steps the quiet row under the box no longer takes (composerRowFit.ts paneWidth). Split it: the quiet row (the default plain footer) keeps "Opus 5.5 Medium" and "Auto" from 760 down to 280 px; the in-box row (plain footer off through the host's composer event, as Debug Settings does) keeps every old step, measured at the same widths. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * cmux NEXT composer: Codex's Ultrafast is a checkbox row under Fast mode With the bolt button gone, a speed option's tier past Fast (Codex's Ultrafast) could not be picked. The model button's menu adds an Ultrafast checkbox row under Fast mode when the option has such a tier: checked while it runs; on picks it, off steps back to Fast. A two-state option has no such row. Restores the picker.tierUltrafast string from the merge base. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * cmux NEXT composer: the row past Fast mode reads Ultrafast only for Codex's Ultrafast tier Any other agent's tier past Fast keeps the name the agent gives it (else its value), so the second checkbox row never calls another harness's tier Ultrafast (Codex review of #24151). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?harness=claude&session=01a16027-99fc-73f0-9482-5a4babfe1b0a&host=cosmic-gray * cmux NEXT composer: keep picker.tierFast, which base's / menu Fast row now names The PR dropped the speed tier strings with FastToggle.tsx; base's SlashMenu (Composer.tsx) reads picker.tierFast, so the merged tree failed typecheck. Restored from base unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?harness=claude&session=01a16027-99fc-73f0-9482-5a4babfe1b0a&host=cosmic-gray * cmux NEXT composer: the full picker's Fast mode row is on for every tier past off The full picker read only the option's first tier as on, so with Codex's Ultrafast running it showed Fast mode off and a click picked Fast instead of turning the tier off (Codex review of #24151 at 9da89d3). One helper, fastModeToggle (modelPickerLayout.ts), now gives both Fast mode rows their on/off state: on is any tier past off, a click turns the running tier off, and the full picker names the running tier (Ultrafast). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?harness=claude&session=01a16027-99fc-73f0-9482-5a4babfe1b0a&host=cosmic-gray * cmux NEXT composer test: the quiet row's model name keeps the button's ink, its effort one step back composer-footer-plain and composer-caret-focus required the model name and its effort to share one colour; this PR puts the effort one step back of the name in the quiet row (composerLocation.css, composer-model-quiet.test.ts), so those checks failed on the PR's head only. In the quiet row they now check the name is in the button's muted ink and the effort differs from it; the New Tab stage row (.nt-pickers), which this PR does not restyle, keeps the old same-ink check. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?harness=claude&session=01a16027-99fc-73f0-9482-5a4babfe1b0a&host=cosmic-gray * cmux NEXT composer: the string catalog only drops the speed button's three strings Re-derived from feat-cmux-next's Localizable.xcstrings with composer.speedLabel, composer.speedTooltip and picker.tierStandard deleted, in base's own formatting and key order. The parsed catalog is unchanged; the diff against base is those three entries alone. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?harness=claude&session=01a16027-99fc-73f0-9482-5a4babfe1b0a&host=cosmic-gray * cmux NEXT composer: split; this PR keeps the quiet model button and the quiet row's fit The speed control's move (the bolt inside the model button, FastToggle.tsx's removal, the Ultrafast menu row, fastModeToggle and speedTierName, the three speed strings and composer-fast-toggle's tests) moves to its own follow-up branch, composer-look-codex-bolt-in-model, so each lands under ~800 changed lines. Those files go back to feat-cmux-next's versions here; composerLocation.css keeps base's speed-button rules, composerRowFit.ts base's bolt sentence, and composer-model-quiet.test.ts drops its two bolt tests (they move with the bolt). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?harness=claude&session=01a16027-99fc-73f0-9482-5a4babfe1b0a&host=cosmic-gray * cmux NEXT composer: the quiet row's harness name follows the row's room, as its other labels do Base's new quiet-row harness (ComposerHarness.tsx) hid its name below a 460 px composer by a container query, while this PR's fit takes no width step in the quiet row: the fit counted the hidden 76 px name as shown and shortened the effort with the room free (a 400 px New Tab read "Med"; CI and huge-blue disagreed on which rows did, by font metrics). The quiet row now skips that query, as it skips the effort, model and access ones, so the name shows while it fits and the fit drops it (data-fit-harness) only when the row runs out. Tests, from the row's measured room (need vs room, probed on huge-blue): - reasoning-chip-steps: 400 px sits within ~20 px of the boundary in all three panes, so it checks the form the row's fit chose, whole; 760 (full) and 300 (short) stay exact. - composer-footer-harness (base's): at 410 px the row has ~60 px spare, so every name shows whole. - composer-row-steps / composer-model-quiet: with the harness's name in the row, whole labels hold down to 400 px (410 px composer); 300/280 px (264 px composer) shorten; widths on the boundary (360 px, 324 px composer) are left out. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?harness=claude&session=01a16027-99fc-73f0-9482-5a4babfe1b0a&host=cosmic-gray * fix(cmux-next): refit quiet composer when controls resize Agent-Session: https://cmux.com/deeplink/agent?harness=codex&session=01a166ed-0099-7a32-8c00-7de0ff0c2191&host=cmux16 * cmux NEXT composer test: the quiet row's widths and ring gap follow base's row Base took the harness label out of the quiet row (#24938) and gave the context affordance one compact size with a 2 px margin (composerFooterShared.css), so: the row now keeps every whole label down to a 244 px composer (measured on the fleet: panes 280 and 300 read "Opus 5.5 Medium"), the ring's glyph sits 15 px after the effort, and the shared focus ring is asserted from the CSS because the gallery stage loads no pages/shared/desktop.css. composer-app-font also reads base's new composerFooterShared.css. Agent-Session: https://cmux.com/deeplink/agent?harness=claude&session=01a174e4-2849-7aa2-bd71-289c4ec9e04f&host=huge-blue * fix(cmux-next): the quiet row's width steps must not outrank the fit Review (codex gpt-6-astra xhigh) found it: `.acpmux-composer-bar:not(.acpmux-composer-foot)` added a class, so each container-query step tied the hiding rules above it at (0,4,0) and, coming later, won. At a 370-440 px composer whose measured fit had already chosen the model's mark (data-fit-model="icon"), the step would have brought the family label back. `:where(:not(...))` adds no specificity, so the steps keep losing to the fit, as before this PR. composer-fit-steps now refuses a bare `:not(.acpmux-composer-foot)` in any step (it catches the regression: the guard fails with the bare selector restored). Also, on the reviewer's second point, the focus hairline is now measured instead of only read: the test puts pages/shared/desktop.css in the stage as the app's pane has it and asserts the shared ring draws 1 px at 22%, not 1.5 px at 48%. Increase Contrast stays a CSS-text assertion, since prefers-contrast does not reach the gallery frame. Agent-Session: https://cmux.com/deeplink/agent?harness=claude&session=01a174e4-2849-7aa2-bd71-289c4ec9e04f&host=huge-blue * Scope the quiet row's fit to the plain look, not the row element The quiet row is a look, not an element: ComposerFoot draws the same .acpmux-composer-bar.acpmux-composer-foot in both looks and marks only the plain one data-footer-look="plain". Keying the width-step guards and paneWidth on the class therefore skipped the steps with the plain footer off too, so the stepped row kept whole labels it had no room for and hid the harness mark that look draws. Both now read data-footer-look. Also follows base: the composer names its family through var(--font-ui) (#24661 and the token move), so the font tests read the token instead of a literal system-ui, and the ring keeps base's compact 22px box. Measured in Chromium on huge-blue, these suites: composer-model-quiet, composer-row-steps, composer-fit-steps, reasoning-chip-steps, composer-footer-plain, composer-alignment-centers, composer-app-font, composer-caret-focus. Before this commit, at merge 966ee3e: 46 pass, 17 fail. After: 87 pass, 0 fail. The same suites at base e3134fd: 45 pass, 0 fail, so none of the 17 was a base red. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?id=aid-23d3b9d81891fbbc413e --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Leo's composer ask (2026-10-24):
This PR handles the first half of that ask: the composer footer no longer renders a harness label or its dedicated CSS/component. The model picker remains the harness-selection surface; the existing machine/location selection is untouched. The gallery fixture and browser assertions now verify the footer has only its mode/model controls and that a new chat opens the model picker rail for harness selection.
Gallery before/after (base
feat-cmux-nextvs this head): composer gallery compareValidation:
bun run typecheck(webviews): pass.bun run check:browser-lanesand targetedvp fmt --check: pass.The overlap scan reported neighboring draft PRs #24519 and #24391 on shared composer files; neither changes this footer harness removal.
Takes back #24169: remove the obsolete harness footer label while keeping harness selection in the model picker.
Agent session: big-red
Need help on this PR? Tag
@codesmith-botwith what you need.