[teamleaderleo]

Summary

Typing a newline in the agent chat's composer (Shift-Return, or a soft wrap) made the transcript's scrollbar jump (Leo, 2026-10-20).

Root cause: the bottom stack floats over the transcript's foot, and VirtualTranscript keeps room under the last row equal to the stack's live height (rowsHeight = rows + THREAD_TOP + dockSpace). Every composer line therefore changed the scroller's content height. Scrolled up, the reader's offset and rows stayed put, but scrollHeight grew by a line each time, so the scrollbar's thumb moved and shrank on each newline (and WebKit flashes the overlay scroller on a content resize). The full-height minimap, sized from height - dockSpace, rescaled every mark on each line too.

Fix (App.tsx): the room under the last row follows the stack only while the view follows the latest row. Scrolled up, it stays as it was, so the scroller's offset, height and scrollbar don't change, and neither does the minimap. Back at the latest row, the room is the stack's again and the last row sits above the grown card. A followed view still stays at its end on every frame as the composer grows, as before. Jump to latest, the mask under the stack, and scroll-into-view still use the stack's live height.

Follow-up to #22819 (bottom stack is a no-scroll zone), built on #22272 and #22858.

Changelog

Fixed: In agent chat, typing a newline in the composer while scrolled up no longer moves the transcript's scrollbar.

Testing

• webviews/test/agent-pane-scrollbar-full-height.test.ts has a new case, "a composer that grows leaves a reader scrolled up where they were, scrollbar and all, on every frame". It scrolls up, grows the composer by four Shift-Return lines and a wrapped line, and samples every painted frame: scrollTop, scrollHeight and clientHeight stay unchanged. It then scrolls back to the end and checks that the latest row sits above the grown card. The existing case "a composer that grows keeps the followed end above it on every frame" covers the pinned view (distance from end stays 0 on every frame).
• I ran these on huge-blue (Linux, Playwright Chromium; CMUX_BROWSER_TESTS=1). The new case fails on base (scrollHeight grew on every line) and passes with the fix, twice. Related suites pass with the fix: agent-pane-bottom-stack, chat-column, follow-live, pr-dock, queue-steer, thread-minimap-, thread-scroll-map, transcript-scroll-restore.
• agent-pane-stream-jitter "webkit: scrolling up while a reply streams is never pulled back" fails the same way on base on that host. agent-pane-wheel-pass-through hit a beforeEach timeout on that host on base too.
• tsc --noEmit, vp fmt --check and vp lint pass on the changed files.

Observed / Not run

• Not run on cosmic-gray (too loaded), and not run in the macOS app's WKWebView. CI's browser lane is the check.

Agent session: cosmic-gray

🤖 Generated with Claude Code

---

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

---

Summary by cubic

Fixes agent chat so growing the composer (newlines or soft wraps) no longer moves the transcript's scrollbar when the reader is scrolled up.

• The room kept under the last row for the floating bottom stack was the stack's live height, so every composer line changed the scroller's content height.
• The room now follows the stack only while the view follows the latest row; scrolled up, it stays as it was, leaving the scrollbar and minimap unchanged.
• Back at the latest row, the room is the stack's again and the latest row sits above the grown card; a followed view still stays at its end.

^{Written for commit 5a5ab52. Summary will update on new commits.}

[teamleaderleo]

Owned by projects-9d on cosmic-gray · session

[github-actions]

cmux status _{(one comment, edited in place; each line is updated by the job that owns it)}

Review 5a5ab52: HQ feed card · gallery before/after runs · checks

App of this head (app.zip, kept 1 day)

gh workflow run reload-build.yml -R manaflow-ai/cmux --ref feat-cmux-next -f tag=pr23097 -f ref=cmux/composer-grow-keeps-scroll -f platform=macos -f backend_mode=local

Unclear: web / react-apps-check failed: no known signature; failed step: Check webview test shards.

CI failed on 5a5ab52dcc (run 39561859342 attempt 1): 4 code, 1 unknown.

Job	Verdict	Why
web / react-apps-check (2)	code	a web test failed
web / react-apps-check (3)	code	a web test failed
web / react-apps-check (4)	code	a web test failed
web / react-apps-check (1)	code	a web test failed
web / react-apps-check	unknown	no known signature; failed step: Check webview test shards

Failures:

• seen on other PRs agent-pane-queue-steer.test.ts queued and steering prompts > the queue takes its own place in the stack: nothing is cut off or draw...: failed (job)
• seen on other PRs composer-location-strip.test.ts the chat composer at 420 px: the strip's marks line up with the prompt's edges: failed (job)
• seen on other PRs composer-location-strip.test.ts the chat composer at 760 px: the strip's marks line up with the prompt's edges: failed (job)
• seen on other PRs composer-mention.test.ts @ then letters lists the chat's matching files and Return writes the path: failed (job)
• seen on other PRs composer-mention.test.ts after picking a path markdown escapes, typing on leaves the picker closed: failed (job)
• seen on other PRs composer-mention.test.ts @proje lists other chats, then folders, then the chat folder's files: failed (job)
• seen on other PRs composer-mention.test.ts a chat named for a long prompt does not take @app from the file: failed (job)
• seen on other PRs composer-mention.test.ts a bare @ lists the most recent other chats: failed (job)
• seen on other PRs composer-mention.test.ts an @ inside a word opens nothing and asks the host for nothing: failed (job)
• seen on other PRs composer-suggested-prompt.test.tsx nothing is suggested from another folder, another worktree, a merged PR, an old chat or a busy one: failed (job)
• ... 13 more in the job logs

Matched log lines

web / react-apps-check (2): (fail) queued and steering prompts > the queue takes its own place in the stack: nothing is cut off or drawn over [1288.62ms]
web / react-apps-check (3): (fail) @ then letters lists the chat's matching files and Return writes the path [2263.59ms]
web / react-apps-check (4): (fail) nothing is suggested from another folder, another worktree, a merged PR, an old chat or a busy one [104.37ms]
web / react-apps-check (1): (fail) with no running call, Ctrl-B stays the composer's and asks nothing [1037.33ms]

Not re-run automatically: web / react-apps-check (2), web / react-apps-check (3), web / react-apps-check (4), web / react-apps-check (1), web / react-apps-check are not machine failures.

Written by scripts/ci/classify_failures.py (ci-failure-attribution.yml); signatures are its SIGNATURES table. A machine verdict is the runner's fault, not this PR's; yours means the failing file is one this PR changes, also red on main that main's latest full suite fails the same way, seen on other PRs that it failed on another pull request's run lately.

red: web / Web status; NEW: green on the base 3d944c5; do not merge until green; see lane-rules
red: web / react-apps-check; no finished base run on 3d944c5 to compare; do not merge until green; see lane-rules
red: web / react-apps-check (1); no finished base run on 3d944c5 to compare; do not merge until green; see lane-rules
red: web / react-apps-check (2); no finished base run on 3d944c5 to compare; do not merge until green; see lane-rules
red: web / react-apps-check (3); no finished base run on 3d944c5 to compare; do not merge until green; see lane-rules
red: web / react-apps-check (4); no finished base run on 3d944c5 to compare; do not merge until green; see lane-rules

CI fast guards failed on 5a5ab52dcc (https://github.com/manaflow-ai/cmux/actions/runs/39561859342). It does not block the merge; a red guard merged into main breaks it for every open PR.
The log named no failed step; see the run.

Agents: python3 scripts/ci/guard_attribution.py fix applies the mechanical fixes locally.

Tour of 5a5ab52dcc (run): no media. The tour did not run: only webview code changed (2 path(s)), which the gallery advisory run checks in Linux Chromium on the pull request (its artifact holds any captures); add the media-tour-mac label to tour it on a Mac.

[teamleaderleo]

Bead cx-lpn2o created for this PR by the cmux coordinator.

[blacksmith-sh]

Found 20 test failures on Blacksmith runners:

Failures

Test	View Logs
test/agent-pane-ctrl-b-background.test.ts/agent-pane-ctrl-b-background.test	View Logs
test/agent-pane-ctrl-b-background.test.ts/agent-pane-ctrl-b-background.test	View Logs
test/agent-pane-peer-chat.test.ts/agent-pane-peer-chat.test	View Logs
test/agent-pane-peer-chat.test.ts/agent-pane-peer-chat.test	View Logs
test/agent-pane-queue-steer.test.ts/agent-pane-queue-steer.test	View Logs
test/agent-pane-queue-steer.test.ts/agent-pane-queue-steer.test	View Logs
test/agent-pane-working-counts.test.ts/agent-pane-working-counts.test	View Logs
test/agent-pane-working-counts.test.ts/agent-pane-working-counts.test	View Logs
test/composer-history.test.ts/composer-history.test	View Logs
test/composer-history.test.ts/composer-history.test	View Logs

...and 10 more test failures. View all on Blacksmith

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

[teamleaderleo]

merge-audit: #23097 merged at 5a5ab52 without no SHIP review receipt on the head. #23244 reverts it. To keep this change, add the missing evidence for 5a5ab52 here; the next audit pass then closes #23244. Otherwise review and land #23244.

[teamleaderleo]

Review (subagent, correctness first): SHIP at 5a5ab52

An independent review subagent read this PR's change at ef6af37 (code only, no local run) and passed it with nothing blocking. 5a5ab52 is a clean merge of origin/feat-cmux-next into that commit: the PR's diff against base is unchanged, still the same two files.

• room reads the same following state as the stick-to-bottom code. While pinned, the path is unchanged, so no frame leaves the view off the end.
• Every way back to the end catches up: wheel, keys, minimap, Find and Jump to latest.
• There is no feedback loop between the room and the pinned state.
• Docked layouts, Quick Chat and replays have dockSpace 0, so they are unchanged.
• The new per-frame test fails on base and passes with the fix, on huge-blue and on CI.
• Non-blocking:
  • after the stack grows while the reader is scrolled up, a long Jump to latest snaps instead of gliding
  • scrolling back to the end can show one frame with the last row under the card
  • the room starts at 0 before the stack is first measured

Risk tier: risk:normal. The change is webviews only, so no Swift or Rust compile applies. Every red at this head was also red on base or on other PRs:

• peer-chat and working-counts: base broke them in cmux-next agent pane: a running call is named once, not on its row and the working line #23090, and test(acpmux): peer-chat and working-counts read the working line #23090 draws #23164 fixed them after this branch's base merge
• Ctrl-B from cmux NEXT: Ctrl-B runs work in the background; Summary shows live background tasks #23044
• the composer-location-strip marks
• defaults-snapshot
• cmux-vm-schema-gate

This receipt should have been posted before the merge; posting it now so merge-audit can close #23244.

[github-actions]

Shipped in:

• cmux NEXT nightly 1.0.0-nightly.3958372135301 (bc661dae0e)