[teamleaderleo]

HOLD: this head still carries the per-row Stop button in SubagentTray.tsx; it lands after that fix is merged in.

Leo, 2026-10-30, on the agent pane's bottom stack:

"the stupid right indent misalignemnt"

This follows #28533 and is stacked on its head e61c418 (#28533 hasn't landed yet), so its diff shows #28533's commits until that lands. It implements ui-think-tank's audit of Leo's bottom-stack complaints:

One column, one gap, nothing removed: every part over the composer (question, notice, PR control, recap, running tray and its header, agents pill, queue) shares the composer box's left and right edges at every width (including the narrow outset and --cv-inset-end) and the box's radius token. The stack owns spacing: one gap token between parts; no part has its own outer margin or bottom padding; the last part sits directly on the box. Leading mark centred 22px in; trailing mark 22px from the right; an empty trailing slot never pushes the last mark inward.

#28533's "N PRs" design is unchanged: an inline control in the quiet row that opens the chips (Leo: "DON'T … TAKE THE ENTIRE ROW").

What changed

1. The tray's trailing grid is carry of #27602: compact waiting messages and subagent tray grid #28113's, taken whole. Base's carry of #27602: compact waiting messages and subagent tray grid #28113 (Leo, 2026-10-29: "There's this right indent for no reason.") already put the header's, every line's and View all's chevrons on one x, with a line's Stop after the chevron. This PR merged base and kept that grid as the settled design: its SubagentTray.tsx, the tray's trailing-track CSS and its tray tests are base's, unchanged. The 22/42 tray grid an earlier head of this PR had is dropped (design/running-tray.md records it under Replaced). agent-pane-stack-edges still checks 22/42 for the other parts and only records the tray's trailing marks; subagent-tray-columns checks its one chevron x.

2. The narrow outset. At composer columns under 298px, the box and its quiet row reach 8px past the column on each side (composerLocation.css). Every stack part now does the same.

   • --acpmux-stack-start and --acpmux-stack-end on the dock write out the same step, because the composer's container query can't reach its siblings.
   • Before, the tray stood 8px inside the box on both sides at 280px; now it's 0.

3. Radius. The tray (10px), the opened PR chips and the recap's card (12px) now take the box's radius token, --acpmux-composer-radius, which is now also defined on the dock. That gives 11px in the Claude look and 18px otherwise.

4. Own margins and padding. These parts take the stack's edges and drop their own bottom padding; one dock rule now gives --acpmux-stack-gap between parts:

   • the recap (summary.css);
   • the legacy PR bars (gallery only);
   • the host's line (styles.css, 8px bottom padding);
   • the agents pill (bottomStack.css, 8px bottom padding).

   The PR fallback row and the in-transcript tray (Labs) now take --cv-inset-end on the right, and the in-transcript tray uses the stack token instead of its 8px margin. The recap's card joins the glass the same way the tray does.

Measurements (test/agent-pane-stack-edges.test.ts, gallery, Chromium)

	Before (#28533 head)	After
Tray chevrons (header, lines), from the right	47px (1100) / 43px (280), one x	47 / 43px, one x (#28113's grid, unchanged)
Tray left / right off the box at 280px	+8 / −8px	0 / 0
Opened PRs left / right off the box at 280px	+8 / −8px	0 / 0
Recap left / right off the box at 280px	+8 / −8px	0 / 0
Tray radius vs the box's (Claude look)	10 vs 11	11 = 11
Recap card radius vs the box's	12 vs 11	11 = 11
Gap between parts, and under the last part	0	0 (token 0)

Each "before" figure was measured on #28533's head e2d5716 (run 41146240957; also run 41144611936). The "after" figures come from the head 3360d26 (run 41172778670, stack-edges.json).

Stills (2x, light and dark, synthetic data)

Each image is one 2x2 grid: top row before (#28533's head), bottom row after (3360d26; its tray lines show #28113's subagent icon from base); left light, right dark. Ten images in all.

Viewed after image: https://raw.githubusercontent.com/manaflow-ai/cmux/pr-media/28587/ux-bs4-recap-1100.png (and every grid below: the recap card, the tray, the opened PRs and the chevron close-ups all stand on the box's edges and radius in the bottom row; at 280 the before row shows the 8px inset).

Recap, 1100 (its card's radius and edges)

Recap, 280 (the card on the box's narrow outset)

Right-edge chevrons close-up, 1100

Right-edge chevrons close-up, 280

Tray expanded, 1100

Tray expanded, 280

Tray collapsed, 1100

Tray collapsed, 280

PR control opened, 1100 (first row hovered)

PR control opened, 280 (first row hovered)

Tests

• New: test/agent-pane-stack-edges.test.ts, committed before the fix. It runs on the agent-pane-column-edges harness at 1100, 420 and 280px.
  • It measures every .acpmux-dock > :not(.acpmux-composer) part plus .acpmux-composer-queue. Each one's left and right edges and radius must equal .acpmux-composer-box's within 0.5px, and the gap under each must be the stack token, or 0 under the last part, which sits on the box.
  • In each of its rows, the last trailing glyph must centre 22px from the right, with a mark before it on 42px; the running tray's rows are recorded only (carry of #27602: compact waiting messages and subagent tray grid #28113's grid).
  • Frames: agent-pane.subagent-tray (between-turns, which is the "Running in background" between-turns state; several-running; duration-columns; panel-two; and the 300px narrow short-dock), agent-pane.bottom-stack (pr-queue, trusted-permission-queue, compact-trust-queue, short-400-long-draft), agent-pane.pr-dock (inline-pills, full-bars), agent-pane.recap-card, and agent-pane.bottom-stack-styles (collapsed, expanded, prs-open).
  • Fail-level for the parts fixed here: the tray, the opened PRs, the count's host, the recap, the agents pill and the host's line. Warning-level for the rest (grid.md §7).
  • Red at e2d5716 with 37 failures (run 41144611936 above); 0 failures at 7eadc7c and at the head 3360d26.
• Updated tests:
  • subagent-tray-columns and subagent-tray are base's (carry of #27602: compact waiting messages and subagent tray grid #28113's), unchanged.
  • subagent-tray-layout checks that the tray's margin is the stack's insets.
• Green: 133 of 134 at 7eadc7c (run 41148834269) across this test, agent-pane-bottom-stack-edges, agent-pane-pr-stack-gate, subagent-tray-columns, subagent-tray, subagent-tray-layout, agent-pane-column-edges, agent-pane-dock-gutters, agent-pane-bottom-stack, composer-bottom-space, composer-edges and pr-dock-inline-layout. The one red was subagent-tray-layout still asserting the old margin literal. It was fixed at 1b674d0 (49 of 49, run 41161006984). At c86b0e6 (rebased onto Agent pane: the bottom stack on one edge as one glass, Stop in the tray row's menu #28533's e61c418) agent-pane-stack-edges, agent-pane-bottom-stack-edges, agent-pane-pr-stack-gate, subagent-tray-layout, subagent-tray-columns and subagent-tray pass 56 of 56 (run 41170446218); at the head 3360d26, with base merged and carry of #27602: compact waiting messages and subagent tray grid #28113's tray tests, the same six files pass 56 of 56 (run 41172778670).
• scripts/web-lint.sh and prepush-gate.py pass. The design-token baseline went down by one entry: the recap's 6px bottom padding is gone.

Still at warning level (not fixed here)

• The permission and question card (.acpmux-permission): radius 8 in trusted-permission-queue, and at 280px it stands +8/−8px inside the box's narrow outset.
• The legacy PR bars (gallery comparison only): content-width bars.
• The leading mark: the tray's icon centres 15px in, not 22px. Moving it means moving the tray's text column, which the running-tray spec and fix(agent-pane): keep running tray durations on one line #28133's grid own; it's left for a follow-up.

Observed / Not run

• Observed: the runs above. The geometry numbers come from stack-edges.json in those runs' artifacts.
• Not run: a native build or a Mac capture. This is web only, and the stills come from the gallery.
• Not run: an independent review at this head.

Spec: design/composer.md rule 9 now carries this rule, and design/running-tray.md records #28113's one trailing track as settled (the 22/42 tray grid under Replaced), both in leo-workspace#1751 (lane/ux-bottom-stack-spec). This PR doesn't duplicate them.

Risk: normal. The change is CSS only (edges, radius, gap and outset); the tray's markup and trailing grid are base's.

Changelog

Fixed: the running tray's chevrons line up with the composer's right edge, and every part over the composer shares its edges and corner radius at every width.

Agent session: agent session

🤖 Generated with Claude Code

[blacksmith-sh]

Found 2 test failures on Blacksmith runners:

Failures

Test	View Logs
test_no_runner_picker_returns_a_hosted_mac (main.NoHostedMacInNightlies.test_no_run ner_picker_returns_a_hosted_mac)/ test_no_runner_picker_returns_a_hosted_mac (main.NoHostedMacInNightlies.test_no_run ner_picker_returns_a_hosted_mac)	View Logs
test_the_repository_highlights_validate (main.DigestTests.test_the_repository_highl ights_validate)/ test_the_repository_highlights_validate (main.DigestTests.test_the_repository_highl ights_validate)	View Logs

^{Need help on this PR? Tag @codesmith-bot with what you need.}

[teamleaderleo]

MODE-03 / UX-12 source-pinned visual audit: https://github.com/manaflow-ai/cmuxterm-hq/issues/5457#issuecomment-6536782475

At feat-cmux-next@ae8cd21c3ef5b3e8b215fe73e7e0fbf6121fe553, bottomStack.css:53-63 still gives the agents-pill wrapper and running tray separate edge treatment. This follow-on PR already owns the geometry fix and its agent-pane-stack-edges.test.ts; #28533 owns the compact N PRs control and one-glass presentation. Suggested acceptance capture once this stack is ready: 1100/420/280px, light/dark and 125% zoom, with busy tray + 2 PRs and a short pane; compare every part's edges within 0.5px, ensure the collapsed count adds no full-width blank row and the composer stays reachable. Source-only observation; no signed build or test run in this chat. No new competing branch requested.

[teamleaderleo]

Owned by an agent session · took over from an earlier session · session

[teamleaderleo]

Owned by an agent session · session

[teamleaderleo]

Review (codex gpt-6.1-sol high fast, router-review): SHIP at df31d42

Follow-ups:

1. P2, defer: webviews/src/agent-session/acpmux/bottomStack.css:67: the stack's outset switches at 297.5px while the composer's container query switches at 297px; at a fractional column (e.g. 297.25px) stack parts expand 8px each side while the composer stays inset. Match the composer's breakpoint exactly.

[teamleaderleo]

drain-ship: not landing. After the lander's base merge, the head (df31d42) still fails cmux-next checks / Default cursor (no pointing hand in app UI), which is green on base, so it is this PR's own red: webviews/src/agent-session/acpmux/bottomStack.css:191 webviews/src/agent-session/acpmux/bottomStack.css:262 add cursor: pointer. Fix: drop the cursor rule (the arrow is the app default), then it needs a fresh review at the new head. Run: https://github.com/manaflow-ai/cmux/actions/runs/41182859775

[teamleaderleo]

Orphaned: no activity for 13 h, and no lane holds a lease on it. The orphan sweep could not merge feat-cmux-next in: conflicts in webviews/scripts/design-tokens-baseline.jsonl, webviews/src/agent-session/acpmux/App.tsx, webviews/src/agent-session/acpmux/bottomStack.css, webviews/src/agent-session/acpmux/composerLocation.css, webviews/src/agent-session/acpmux/summary/summary.css. Listed for a lane to resolve.

[teamleaderleo]

Awareness note from an agent lane that does not own this PR; it is not a review or a request to take the PR over.

---

Consolidating the still-needed running-tray changes in #29794, preserving commit authorship and the newer tray layout already on base. This keeps the tray alignment, folding and hidden-row fixes together for validation and landing.

[cubic-dev-ai]

10 issues found across 16 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="webviews/src/agent-session/acpmux/bottomStackStyles.gallery.ts">

<violation number="1" location="webviews/src/agent-session/acpmux/bottomStackStyles.gallery.ts:64">
P3: The `no-prs` variants still display “Both PRs are open” even though `prs: false` removes both pull-request activities. Branch the assistant fixture text on `prs` as well, so these gallery states describe the data they render.</violation>
</file>

<file name="webviews/src/agent-session/acpmux/Localizable.xcstrings">

<violation number="1" location="webviews/src/agent-session/acpmux/Localizable.xcstrings:16388">
P2: These count entries are English in every non-English locale, so localized users see `1 PR`/`{count} PRs` even though the surrounding dock label is translated. Add locale-specific count wording, including each locale’s plural form, instead of duplicating the English value across all locale entries.</violation>
</file>

<file name="webviews/test/agent-pane-pr-stack-gate.test.ts">

<violation number="1" location="webviews/test/agent-pane-pr-stack-gate.test.ts:67">
P2: This gate ignores every JSX-attribute use of `composer`, not just the known `QuickSurface` prop, so a later unguarded composer prop could bypass the Sandboxed visibility check while the test still passes. Restrict the exclusion to the verified `QuickSurface` attribute (or assert that all excluded uses are exactly that one).</violation>
</file>

<file name="webviews/test/agent-pane-stack-edges.test.ts">

<violation number="1" location="webviews/test/agent-pane-stack-edges.test.ts:103">
P2: The width matrix is not exercising its requested widths because `open` never passes `width` to `galleryFrame`; the gallery defaults to its `normal` width. Pass `width` in the frame parameters so the 420px and 280px narrow-screen assertions actually run against narrow layouts.</violation>

<violation number="2" location="webviews/test/agent-pane-stack-edges.test.ts:159">
P2: This test can pass vacuously when a target part is not rendered, because it derives the parts under test from the DOM instead of asserting expected parts per fixture. Add fixture expectations and fail when a required part is missing, especially for the parts in `FIXED`.</violation>
</file>

<file name="webviews/test/agent-pane-bottom-stack-edges.test.ts">

<violation number="1" location="webviews/test/agent-pane-bottom-stack-edges.test.ts:44">
P2: This matrix never exercises the narrow outset: 420px is above the `max-width: 297px` breakpoint. Include 280px so the new edge and chip-width assertions cover the narrow layout.</violation>
</file>

<file name="webviews/src/agent-session/acpmux/summary/PullRequestStack.tsx">

<violation number="1" location="webviews/src/agent-session/acpmux/summary/PullRequestStack.tsx:45">
P2: This reverses first-seen order instead of newest-touch order, so mutating an older pull request leaves its chip behind newer untouched requests. Sort the visible results by `touched` (or reuse the Changes ordering) before rendering the stack.</violation>
</file>

<file name="webviews/src/agent-session/acpmux/bottomStack.css">

<violation number="1" location="webviews/src/agent-session/acpmux/bottomStack.css:210">
P2: This `pointer-events: none` declaration is overridden in the floating dock by the more-specific direct-child rule, so the `above` count host captures clicks across its full row rather than only its button. Increase this selector’s specificity or otherwise ensure the host remains pointer-transparent.</violation>
</file>

<file name=".github/design-surfaces.json">

<violation number="1" location=".github/design-surfaces.json:42">
P2: This new composer mapping points to a file that does not exist, so the design-surface validity test fails and the map emits a stale-path warning. Remove this entry; `PullRequestStack.tsx` is already mapped.</violation>
</file>

<file name="webviews/src/agent-session/acpmux/subagents/subagents.css">

<violation number="1" location="webviews/src/agent-session/acpmux/subagents/subagents.css:839">
P2: The Labs `after-latest-message` tray is outside `.acpmux-dock`, so `--acpmux-stack-gap` is undefined and this change removes its existing 8px bottom spacing. Preserve the 8px gap for this transcript placement, or define the shared token on an ancestor that also contains this tray.</violation>
</file>

<sub>Reply with feedback, questions, or to request a fix.

View guided diff | Turn on auto-fix | Re-trigger cubic</sub>

[cubic-dev-ai]

P2: These count entries are English in every non-English locale, so localized users see 1 PR/{count} PRs even though the surrounding dock label is translated. Add locale-specific count wording, including each locale’s plural form, instead of duplicating the English value across all locale entries.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At webviews/src/agent-session/acpmux/Localizable.xcstrings, line 16388:

<comment>These count entries are English in every non-English locale, so localized users see `1 PR`/`{count} PRs` even though the surrounding dock label is translated. Add locale-specific count wording, including each locale’s plural form, instead of duplicating the English value across all locale entries.</comment>

<file context>
@@ -16252,6 +16252,140 @@
+      }
+    },
+    "prDock.countOne": {"comment":"Agent pane: compact count of one pull request above the composer (si-prpills)","extractionState":"manual","localizations":{"en":{"stringUnit":{"state":"needs_review","value":"1 PR"}},"ar":{"stringUnit":{"state":"needs_review","value":"1 PR"}},"bs":{"stringUnit":{"state":"needs_review","value":"1 PR"}},"da":{"stringUnit":{"state":"needs_review","value":"1 PR"}},"de":{"stringUnit":{"state":"needs_review","value":"1 PR"}},"es":{"stringUnit":{"state":"needs_review","value":"1 PR"}},"fr":{"stringUnit":{"state":"needs_review","value":"1 PR"}},"it":{"stringUnit":{"state":"needs_review","value":"1 PR"}},"ja":{"stringUnit":{"state":"needs_review","value":"1 PR"}},"km":{"stringUnit":{"state":"needs_review","value":"1 PR"}},"ko":{"stringUnit":{"state":"needs_review","value":"1 PR"}},"nb":{"stringUnit":{"state":"needs_review","value":"1 PR"}},"pl":{"stringUnit":{"state":"needs_review","value":"1 PR"}},"pt-BR":{"stringUnit":{"state":"needs_review","value":"1 PR"}},"ru":{"stringUnit":{"state":"needs_review","value":"1 PR"}},"th":{"stringUnit":{"state":"needs_review","value":"1 PR"}},"tr":{"stringUnit":{"state":"needs_review","value":"1 PR"}},"uk":{"stringUnit":{"state":"needs_review","value":"1 PR"}},"vi":{"stringUnit":{"state":"needs_review","value":"1 PR"}},"zh-Hans":{"stringUnit":{"state":"needs_review","value":"1 PR"}},"zh-Hant":{"stringUnit":{"state":"needs_review","value":"1 PR"}}}},
+    "prDock.count": {"comment":"Agent pane: compact count of pull requests above the composer; {count} is the number (si-prpills)","extractionState":"manual","localizations":{"en":{"stringUnit":{"state":"needs_review","value":"{count} PRs"}},"ar":{"stringUnit":{"state":"needs_review","value":"{count} PRs"}},"bs":{"stringUnit":{"state":"needs_review","value":"{count} PRs"}},"da":{"stringUnit":{"state":"needs_review","value":"{count} PRs"}},"de":{"stringUnit":{"state":"needs_review","value":"{count} PRs"}},"es":{"stringUnit":{"state":"needs_review","value":"{count} PRs"}},"fr":{"stringUnit":{"state":"needs_review","value":"{count} PRs"}},"it":{"stringUnit":{"state":"needs_review","value":"{count} PRs"}},"ja":{"stringUnit":{"state":"needs_review","value":"{count} PRs"}},"km":{"stringUnit":{"state":"needs_review","value":"{count} PRs"}},"ko":{"stringUnit":{"state":"needs_review","value":"{count} PRs"}},"nb":{"stringUnit":{"state":"needs_review","value":"{count} PRs"}},"pl":{"stringUnit":{"state":"needs_review","value":"{count} PRs"}},"pt-BR":{"stringUnit":{"state":"needs_review","value":"{count} PRs"}},"ru":{"stringUnit":{"state":"needs_review","value":"{count} PRs"}},"th":{"stringUnit":{"state":"needs_review","value":"{count} PRs"}},"tr":{"stringUnit":{"state":"needs_review","value":"{count} PRs"}},"uk":{"stringUnit":{"state":"needs_review","value":"{count} PRs"}},"vi":{"stringUnit":{"state":"needs_review","value":"{count} PRs"}},"zh-Hans":{"stringUnit":{"state":"needs_review","value":"{count} PRs"}},"zh-Hant":{"stringUnit":{"state":"needs_review","value":"{count} PRs"}}}},
     "prDock.dismiss": {
       "comment": "Agent pane: button that hides a pull request's bar above the composer for this chat; {number} is the PR number (cx-ncc.81)",
</file context>

[cubic-dev-ai]

P2: This gate ignores every JSX-attribute use of composer, not just the known QuickSurface prop, so a later unguarded composer prop could bypass the Sandboxed visibility check while the test still passes. Restrict the exclusion to the verified QuickSurface attribute (or assert that all excluded uses are exactly that one).

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At webviews/test/agent-pane-pr-stack-gate.test.ts, line 67:

<comment>This gate ignores every JSX-attribute use of `composer`, not just the known `QuickSurface` prop, so a later unguarded composer prop could bypass the Sandboxed visibility check while the test still passes. Restrict the exclusion to the verified `QuickSurface` attribute (or assert that all excluded uses are exactly that one).</comment>

<file context>
@@ -0,0 +1,85 @@
+      while (node && !ts.isJsxExpression(node)) node = node.parent;
+      return node as ts.JsxExpression | undefined;
+    })
+    .filter((use): use is ts.JsxExpression => use !== undefined && !ts.isJsxAttribute(use.parent));
+  expect(uses.length).toBeGreaterThan(0);
+  for (const use of uses) {
</file context>

[cubic-dev-ai]

P2: This test can pass vacuously when a target part is not rendered, because it derives the parts under test from the DOM instead of asserting expected parts per fixture. Add fixture expectations and fail when a required part is missing, especially for the parts in FIXED.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At webviews/test/agent-pane-stack-edges.test.ts, line 159:

<comment>This test can pass vacuously when a target part is not rendered, because it derives the parts under test from the DOM instead of asserting expected parts per fixture. Add fixture expectations and fail when a required part is missing, especially for the parts in `FIXED`.</comment>

<file context>
@@ -0,0 +1,319 @@
+      return fill || s.boxShadow !== "none" || parseFloat(s.borderTopWidth) > 0 || parseFloat(s.borderLeftWidth) > 0;
+    };
+    const composer = dock.querySelector(":scope > .acpmux-composer");
+    const nodes = [...dock.children].filter((el) => el !== composer && shown(el));
+    const queue = composer?.querySelector(":scope > .acpmux-composer-queue");
+    if (queue && shown(queue)) nodes.push(queue);
</file context>

[cubic-dev-ai]

P2: The width matrix is not exercising its requested widths because open never passes width to galleryFrame; the gallery defaults to its normal width. Pass width in the frame parameters so the 420px and 280px narrow-screen assertions actually run against narrow layouts.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At webviews/test/agent-pane-stack-edges.test.ts, line 103:

<comment>The width matrix is not exercising its requested widths because `open` never passes `width` to `galleryFrame`; the gallery defaults to its `normal` width. Pass `width` in the frame parameters so the 420px and 280px narrow-screen assertions actually run against narrow layouts.</comment>

<file context>
@@ -0,0 +1,319 @@
+  async function open(entry: string, variant: string, width: number, theme?: string, scale = 1): Promise<Page> {
+    const port = (vite!.httpServer!.address() as { port: number }).port;
+    const page = await browser!.newPage({ viewport: { width, height: 680 }, deviceScaleFactor: scale });
+    await page.goto(`http://127.0.0.1:${port}${galleryFrame(entry, { variant, theme })}`);
+    await page.locator(".acpmux-composer-box").first().waitFor({ timeout: 90_000 });
+    await page.waitForFunction(() => document.documentElement.dataset.galleryReady === "1", undefined, {
</file context>

Suggested change

	await page.goto(`http://127.0.0.1:${port}${galleryFrame(entry, { variant, theme })}`);
	await page.goto(`http://127.0.0.1:${port}${galleryFrame(entry, { variant, theme, width })}`);

[cubic-dev-ai]

P2: This matrix never exercises the narrow outset: 420px is above the max-width: 297px breakpoint. Include 280px so the new edge and chip-width assertions cover the narrow layout.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At webviews/test/agent-pane-bottom-stack-edges.test.ts, line 44:

<comment>This matrix never exercises the narrow outset: 420px is above the `max-width: 297px` breakpoint. Include 280px so the new edge and chip-width assertions cover the narrow layout.</comment>

<file context>
@@ -0,0 +1,361 @@
+  setDefaultTimeout(840_000);
+  const ENTRY = "agent-pane.bottom-stack-styles";
+  const out = process.env.CMUX_BOTTOM_STACK_CAPTURE ?? process.env.FAST_CHECKS_OUT;
+  const WIDTHS = [1100, 760, 420];
+  /** The selected look's states, held to every rule; the comparisons are measured for the record
+   *  and held to the no-row rule where they claim it. */
</file context>

[cubic-dev-ai]

P2: This reverses first-seen order instead of newest-touch order, so mutating an older pull request leaves its chip behind newer untouched requests. Sort the visible results by touched (or reuse the Changes ordering) before rendering the stack.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At webviews/src/agent-session/acpmux/summary/PullRequestStack.tsx, line 45:

<comment>This reverses first-seen order instead of newest-touch order, so mutating an older pull request leaves its chip behind newer untouched requests. Sort the visible results by `touched` (or reuse the Changes ordering) before rendering the stack.</comment>

<file context>
@@ -0,0 +1,291 @@
+    parsed.current = { key, repository, prs: transcriptPullRequests(rows, { folder: repository, replies: false }) };
+  }
+  const all = parsed.current.prs;
+  return useMemo(() => visibleChanges(all, headerStatus(live)).reverse(), [all, live]);
+}
+
</file context>

[cubic-dev-ai]

P2: This pointer-events: none declaration is overridden in the floating dock by the more-specific direct-child rule, so the above count host captures clicks across its full row rather than only its button. Increase this selector’s specificity or otherwise ensure the host remains pointer-transparent.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At webviews/src/agent-session/acpmux/bottomStack.css, line 210:

<comment>This `pointer-events: none` declaration is overridden in the floating dock by the more-specific direct-child rule, so the `above` count host captures clicks across its full row rather than only its button. Increase this selector’s specificity or otherwise ensure the host remains pointer-transparent.</comment>

<file context>
@@ -49,10 +49,265 @@
+  position: relative;
+  flex: none;
+  margin: 0 var(--acpmux-stack-end) 0 var(--acpmux-stack-start);
+  pointer-events: none;
+}
+.acpmux-dock > .acpmux-pr-count-host > .acpmux-pr-count {
</file context>

Suggested change

	pointer-events: none;
	pointer-events: none !important;

[cubic-dev-ai]

P2: This new composer mapping points to a file that does not exist, so the design-surface validity test fails and the map emits a stale-path warning. Remove this entry; PullRequestStack.tsx is already mapped.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At .github/design-surfaces.json, line 42:

<comment>This new composer mapping points to a file that does not exist, so the design-surface validity test fails and the map emits a stale-path warning. Remove this entry; `PullRequestStack.tsx` is already mapped.</comment>

<file context>
@@ -36,7 +36,18 @@
+        "webviews/src/agent-session/acpmux/composerStates.css",
+        "webviews/src/agent-session/acpmux/bottomStack.css",
+        "webviews/src/agent-session/acpmux/summary/PullRequestStack.tsx",
+        "webviews/src/agent-session/acpmux/summary/PullRequestDockLegacy.tsx"
+      ]
+    },
</file context>

[cubic-dev-ai]

P2: The Labs after-latest-message tray is outside .acpmux-dock, so --acpmux-stack-gap is undefined and this change removes its existing 8px bottom spacing. Preserve the 8px gap for this transcript placement, or define the shared token on an ancestor that also contains this tray.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At webviews/src/agent-session/acpmux/subagents/subagents.css, line 839:

<comment>The Labs `after-latest-message` tray is outside `.acpmux-dock`, so `--acpmux-stack-gap` is undefined and this change removes its existing 8px bottom spacing. Preserve the 8px gap for this transcript placement, or define the shared token on an ancestor that also contains this tray.</comment>

<file context>
@@ -830,9 +831,12 @@ button.cv-subagent > :last-child {
 .acpmux-running-tray[data-placement="after-latest-message"] {
   flex: none;
-  margin: 0 var(--cv-inset) 8px;
+  margin: 0 var(--cv-inset-end, var(--cv-inset)) var(--acpmux-stack-gap, 0px) var(--cv-inset);
 }
 .acpmux-agent-card__state[data-status="failed"] {
</file context>

Suggested change

	margin: 0 var(--cv-inset-end, var(--cv-inset)) var(--acpmux-stack-gap, 0px) var(--cv-inset);
	margin: 0 var(--cv-inset-end, var(--cv-inset)) 8px var(--cv-inset);

[cubic-dev-ai]

P3: The no-prs variants still display “Both PRs are open” even though prs: false removes both pull-request activities. Branch the assistant fixture text on prs as well, so these gallery states describe the data they render.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At webviews/src/agent-session/acpmux/bottomStackStyles.gallery.ts, line 64:

<comment>The `no-prs` variants still display “Both PRs are open” even though `prs: false` removes both pull-request activities. Branch the assistant fixture text on `prs` as well, so these gallery states describe the data they render.</comment>

<file context>
@@ -0,0 +1,190 @@
+        ? [activity([create("fix(picker): keep the model list in view", 128), create("Composer: one edge", 129)], 13)]
+        : []),
+      ...(tray ? [row(SUBAGENTS, 12, { subagents: agents })] : []),
+      assistant(
+        tray
+          ? "Both PRs are open. Nine subagents are still checking in the background."
</file context>

Suggested change

	assistant(
	tray
	? "Both PRs are open. Nine subagents are still checking in the background."
	: "Both PRs are open and the checks are done.",
	2,
	),
	],
	assistant(
	prs
	? tray
	? "Both PRs are open. Nine subagents are still checking in the background."
	: "Both PRs are open and the checks are done."
	: tray
	? "Nine subagents are still checking in the background."
	: "The checks are done.",
	2,
	),

[teamleaderleo]

Merge of feat-cmux-next into this head (0baccf8) needs rework, not a conflict fix. Base has since redesigned or taken most of what this PR carries:

• The "N PRs" count and opened chips are gone on base (Agent chat: remove the PR chip row from transcript messages #29614, Leo: "remove the chips they're so bad"; Agent pane: the bottom stack on one edge as one glass, Stop in the tray row's menu #28533 itself landed with "drop the N PRs control"). This branch still carries Agent pane: the bottom stack on one edge as one glass, Stop in the tray row's menu #28533's earlier chip commits, so a merge brings back summary/PullRequestStack.tsx, agent-pane-pr-stack-gate.test.ts, the App.tsx provider wiring and its strings. Base's design should win here, so all of that has to go.
• The tray's edges, narrow outset, radius and gap are on base already: bottomStack.css gives the tray --acpmux-tray-outset (the same 297.5px clamp, citing Agent pane: one column, one gap for the bottom stack (tray chevron on 22, narrow outset, radius token) #28587), subagents.css gives it --acpmux-composer-radius, --acpmux-stack-gap is base's, and base joins the tray to the composer through [data-card] (Running tray: Stop inside its line, one chevron column, the header's time after its counts #29264, fix(tray): align status and controls with the composer #29794/fix(tray): align status and controls with the composer #29885).
• The per-row Stop that the HOLD waits on is base's settled design now. Running tray: Stop inside its line, one chevron column, the header's time after its counts #29264 ("Running tray: Stop inside its line") landed after Agent pane: the bottom stack on one edge as one glass, Stop in the tray row's menu #28533 and keeps the line's Stop button alongside the menu item. So the HOLD's condition won't be met on base, and once this branch takes base's SubagentTray.tsx it no longer carries anything of its own there.
• test/agent-pane-stack-edges.test.ts measures PR chips and rows that no longer exist. Base's agent-pane-bottom-stack-edges.test.ts covers the tray.

What is still this PR's own on top of base is small: the recap (summary.css: no 6px bottom padding, stack edges at the narrow outset, the box's radius token, joining the glass), the host's line (styles.css: no 8px bottom padding, stack edges), and the dock-level --acpmux-stack-start/end vars plus the one-gap rule for those parts. The clean path is a fresh small PR from feat-cmux-next with only those pieces and a focused measurement in base's bottom-stack-edges test, after which this PR is closed in its favour. Merging this branch would undo base's chip removal.

[teamleaderleo]

Carried to #32669 on current feat-cmux-next. Base already covers the tray's edges, narrow outset, radius and gap, and removed the PR chips (#29614), and the per-row Stop this PR's HOLD waited on is now base's settled design (#29264). #32669 ports only what was still this PR's own: the recap card's and the host line's edges (--cv-inset-end plus the narrow outset) and the recap's radius token, with a recap measurement in the bottom-stack edges test.