[teamleaderleo]

Opening or closing anything in the agent chat's transcript no longer moves your view. The line you clicked (a tool group, a tool, command or MCP call, a settled run, a folded thought, a turn's "Worked for") stays on the pixel you clicked it at while what it opens lays out. Before, a chat following its reply jumped to the new latest row as soon as something opened, so the line you clicked shot up by everything it revealed.

Every one of those collapsibles now opens and closes in one motion: the body's height and opacity, easing out, with the chevron turning on the same duration. Under Reduce Motion, or with the app's animations off, it opens and closes at once.

How it holds still. A capture-phase click on a disclosure line (foldAnchor.ts) records where the line is drawn. Until its fold has played (plus 150 ms for the last measure and the rows "Worked for" inserts), each placement pass in VirtualTranscript puts that line back by moving the offset. It does that instead of following the latest row or keeping the top row. When you're following the reply and what opened pushes the latest row out of view, the view stops following (Jump to latest shows), so the reply streaming on doesn't pull you away. There's one case where the line can still move: closing something at the very end of the transcript. There's nothing below to fill the view, so the browser clamps the offset and the rows above come down by what closed, as in any scroller.

The motion. useFold (fold.ts) keeps a closing body drawn until it has slid away. If you reverse a fold midway, it starts from where it is drawn. It reads --cv-fold-duration, which conversation.css sets from the host's move duration (--agent-motion-move, AgentPaneTheme.swift: about 190 ms, and 0 under Reduce Motion or with animation off), with 180 ms when there's no host. The chevron (.cv-rotor) turns on the same token. The newtab-resize-anim lane owns shared duration tokens but hasn't published any yet. I messaged it, and this is the single alias for it to repoint. ThoughtRow drops its own 200 ms height-only animation for the shared fold. Only the reader's toggle plays it; the automatic fold still snaps.

The rows "Worked for" opens are whole virtual rows. They get the transcript's existing row entrance, not a height fold.

Changelog

• Changed: Opening or closing a tool call, a thought or "Worked for" in agent chat animates and keeps the clicked line where it was, instead of jumping the view.

Observed / Not run

Ran locally (bun 1.4.2, webviews/):

• bun test test/transcript-fold-anchor.test.tsx: red on base 7a4b464 (the "Worked for" line moved from 453.5 px to 235.5 px down the view), green on this head (2 pass). The test is committed first (e9418c6).
• bun test test/transcript-fold-motion.test.tsx: 13 pass. Each collapsible plays height and opacity over the token, keeps a closing body until it finishes, and doesn't animate under Reduce Motion, with data-agent-motion="off", or with a 0 ms token. A reversed fold starts from its drawn height, and the CSS ties the token to --agent-motion-move and the chevron.
• ThoughtRow.test.tsx (20 pass), 21 other transcript, find and tool-row jsdom suites (109 pass, 0 fail) and gallery-coverage.test.ts pass.
• bun run check: formatting, layers and type check pass. It exits 1 on one lint warning, jsx-a11y(prefer-tag-over-role) in src/pages/history/HistoryPage.tsx:177. This PR doesn't touch that file, and the warning is the same on base 7f3a43a.

Not run locally (browser tests run only in CI here):

• test/transcript-fold-anchor-browser.test.ts (Chromium). It checks that the clicked line moves at most 1 px on every frame of the fold, while following a streaming reply and for "Worked for", and that no fold plays under Reduce Motion.
• The new gallery play agent-pane.transcript / fold-holds-clicked-line. At the latest row it opens Worked for, then a command, then closes it, and fails if any line moved.
• No app build: the change is webview-only.

Agent session: cosmic-gray

🤖 Generated with Claude Code

---

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

---

Summary by cubic

Opening or closing a tool call, thought, "Worked for", or any other collapsible in agent chat now animates and keeps the clicked line exactly where it was, instead of jumping the view.

• Every collapsible opens and closes with one shared fold: height and opacity over --cv-fold-duration (the host's move duration, ~190 ms), with the chevron turning on the same token.
• Under Reduce Motion or with the pane's animations off, everything opens and closes instantly.
• When the opened content pushes the streaming reply out of view, the view stops following (Jump to latest appears); the line still holds until the fold settles.
• A closing fold stays pinned shut until its body is removed, so the rows below don't jump the frame before React commits.
• One exception: closing something at the very end of the transcript can still shift the view, as the browser clamps the scroll offset with nothing below to fill it.

^{Written for commit 5f7be0d. Summary will update on new commits.}

[cursor]

Bugbot needs on-demand usage enabled

Bugbot uses usage-based billing for this team and requires on-demand usage to be enabled.

A team admin can enable on-demand usage in the Cursor dashboard.

[github-actions]

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

Review 5f7be0d: 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=pr23093 -f ref=chat/collapse-anchor-anim -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 5f7be0d384 (run 39561724754 attempt 1): 4 code, 1 unknown.

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

Failures:

• 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)
• seen on other PRs composer-suggested-prompt.test.tsx a folder with a trailing slash is the same checkout: failed (job)
• seen on other PRs composer-suggested-prompt.test.tsx a chat that knows its branch takes no PR from another branch: failed (job)
• seen on other PRs composer-suggested-prompt.test.tsx a chat with turns gets no suggestion: failed (job)
• seen on other PRs composer-suggested-prompt.test.tsx Tab writes the suggestion without sending; Shift-Tab does not take it; Esc puts it away: failed (job)
• seen on other PRs composer-suggested-prompt.test.tsx the chat folder's PR from the header's gh read is suggested when no session reports one: failed (job)
• seen on other PRs quiet-access-chip.test.ts chromium: Full access draws in the model chip's color: failed (job)
• seen on other PRs quiet-access-chip.test.ts webkit: Full access draws in the model chip's color: failed (job)
• seen on other PRs agent-pane-ctrl-b-background.test.ts with no running call, Ctrl-B stays the composer's and asks nothing: failed (job)
• seen on other PRs agent-pane-ctrl-b-background.test.ts a Codex chat shows no hint and keeps Ctrl-B: failed (job)
• ... 12 more in the job logs

Matched log lines

web / react-apps-check (4): (fail) nothing is suggested from another folder, another worktree, a merged PR, an old chat or a busy one [51.57ms]
web / react-apps-check (1): (fail) with no running call, Ctrl-B stays the composer's and asks nothing [1013.27ms]
web / react-apps-check (3): (fail) @ then letters lists the chat's matching files and Return writes the path [2167.48ms]
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 [1300.85ms]

Not re-run automatically: web / react-apps-check (4), web / react-apps-check (1), web / react-apps-check (3), web / react-apps-check (2), 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 0f6199c; do not merge until green; see lane-rules
red: web / react-apps-check; no finished base run on 0f6199c to compare; do not merge until green; see lane-rules
red: web / react-apps-check (1); no finished base run on 0f6199c to compare; do not merge until green; see lane-rules
red: web / react-apps-check (2); no finished base run on 0f6199c to compare; do not merge until green; see lane-rules
red: web / react-apps-check (3); no finished base run on 0f6199c to compare; do not merge until green; see lane-rules
red: web / react-apps-check (4); no finished base run on 0f6199c to compare; do not merge until green; see lane-rules

CI fast guards failed on 5f7be0d384 (https://github.com/manaflow-ai/cmux/actions/runs/39561724754). 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 5f7be0d384 (run): no media. The tour did not run: only webview code changed (16 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]

Owned by projects-62 on cosmic-gray · session

[teamleaderleo]

Owned by projects-62 on cosmic-gray · session

[teamleaderleo]

Bead cx-oyt23 created for this PR by the cmux coordinator.

[blacksmith-sh]

Found 18 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 8 more test failures. View all on Blacksmith

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

[teamleaderleo]

Review (claude-review-subagent): BLOCK at 66f967e

Blocking

1. webviews/src/agent-session/acpmux/conversation/fold.ts:675-688: every animated close shows a one-frame pop at its end. The close animation has no fill, so when it finishes the box goes straight back to its natural height and opacity: 1. The body is only removed when React commits setClosing(false) from onfinish. A native finish event gives a DefaultLane update, and that commit lands after the frame the event fires in has painted. In that frame, the body is drawn at full height (or, if the row's ResizeObserver flushSync picks up the update, its row is still placed at the full measured height, leaving a gap). The rows below jump by the body's height for a frame and then snap back. I checked this with a standalone page in Playwright Chromium and WebKit: a height and opacity animation to 0 with removal scheduled via MessageChannel from onfinish. The requestAnimationFrame of the finish frame sees the element still connected at h: 200, opacity: 1 in both engines. The old ThoughtRow code had the same pattern, but this PR applies it to every collapsible (tool groups, tool runs, tool rows, command rows, MCP rows, thoughts), which is the motion this PR is about. Fix: play the close with fill: "forwards" (the reopen path already cancels running), or set node.style.height = "0px" and opacity = "0" in onfinish before setClosing(false) and clear them when it opens. Add a jsdom assertion for it (closing options carry fill: "forwards", or the box stays at 0 after finish() until it unmounts).

Notes (non-blocking)

• fold.ts:673: the opening animates to the scrollHeight measured when it opens. Content that loads during the 190ms (highlighted code, EditDiff hunks) jumps when the fold finishes. This is minor.
• .cv-fold { display: flow-root } keeps the bodies' margins (.cv-shell and .cv-tool-output-card have margin: 4px 0 8px) inside the row, where they could collapse out before. A row ending in an opened body is now about 8px taller. This looks intended, but check it in the gallery.
• The anchor and hold logic in App.tsx (the folding placement branch, layoutDriven gated on foldAnchor, capture-phase click hold, release on wheel/touch/non-activating keys) reads correctly. I found no regressions for streaming follow, ended turns, or Find. Find's reveal hold (at least 600ms, tracking the match each frame) outlasts the fold. Virtualized rows re-measure through the ResizeObserver every frame of the fold, so no row version change is needed. No new strings, no Swift, no literal colors. Reduce Motion, data-agent-motion="off" and a 0ms token are all instant and tested.

Ran (scratch worktree at the head)

• bun test test/transcript-fold-motion.test.tsx test/transcript-fold-anchor.test.tsx src/agent-session/acpmux/conversation/ThoughtRow.test.tsx: 35 pass, 0 fail.
• bun test for TurnRows, agent-pane-mcp-tool-rows, tool-row-relative-path, transcript-density, transcript-live-runs and transcript-output-copy: 33 pass, 0 fail.
• bun test for transcript-disclosure, transcript-find-follow, transcript-fold-anchor, transcript-hidden-follow, transcript-follow, chat-find and agent-pane-follow-live: 6 pass, 3 skipped (browser lane), 0 fail.
• bunx tsc --noEmit -p .: clean.
• transcript-fold-anchor-browser.test.ts did not run: the Vite warm-up timed out at 100s with the host at load 99. It is left to CI.

[teamleaderleo]

Review (claude-review-subagent): SHIP at 5f7be0d

Independent review of the whole PR diff against its merge base with feat-cmux-next (0f6199c), with most attention on the fix for the earlier BLOCK (5f7be0d).

Blocking: none.

The close pop (earlier BLOCK) is fixed

• onfinish for a close now pins .cv-fold shut (height 0, opacity 0, overflow hidden) before it calls setClosing(false). In the frame after the fill-less animation lets go and before React's commit, the box stays at 0, so nothing flashes. The pinned box is the .cv-fold that unmounts when present goes false, so none of those inline styles outlive it.
• Reopening in the same frame the close ends: the layout effect calls pin(node, undefined) first. running.current is already cleared, so the open starts from 0/0 and grows to scrollHeight, then clears overflow. The new test (same box, starts at 0px, no inline styles left) covers this.
• Reopening mid-close: turning reads the animated rect height and computed opacity, cancels the old animation (its onfinish doesn't fire, and its identity guard protects it anyway), and animates from there.
• Reduce Motion, data-agent-motion="off" or a 0 ms token: closing is never set at render, or the layout effect clears it synchronously before paint. Either way the pin is cleared first, so no overflow:hidden is left behind.
• Find opening a closing fold: tool, command, MCP, group and run folds take the turning path. A thought snaps (toggled was already reset after the close's layout effect) and its pin is cleared. Find's reveal hold keeps the match in view: the match sits below the fold's fixed top, so the fold growing doesn't move it.

Non-blocking

1. A cancelled fold never settles. applyAgentTheme (shared/theme.ts:87) calls cancel() on every document animation when a motion-off theme arrives. A close caught in flight then never gets onfinish, so closing stays true and running.current stays set. The body stays drawn under a closed chevron, with aria-expanded="false", until the reader toggles it again (that toggle recovers it). An open caught in flight keeps an inline overflow:hidden. This needs the setting changed within about 190 ms of a click, and the old ThoughtRow had the same gap. An animation.oncancel that settles the same way when running.current === animation would close it.
2. An expired foldAnchor is only cleared on the next placement pass. onScroll's !foldAnchor.current treats it as live until then. It's benign in practice: the next layout pass clears it before the scroll events it causes, and a click already makes scrolls non-programmatic for READER_SCROLL_MS. Checking until in onScroll too would make that explicit.

Also checked

• App.tsx hold: the capture-phase click hold runs before the row's handler and finishes any glide on .acpmux-thread (own animations only, not the subtree). Placement passes during the hold only put the clicked line back. Following stops only when the opened body pushes the latest row out of view. Wheel, touch and non-Enter/Space keys release the hold, and keyboard activation still holds. Listeners are removed on cleanup.
• Merge bf2f27b kept base's comment line (wheelPassThrough). No code change.
• Memo/version staleness: fold state is local to each row component. Row height changes reach VirtualTranscript through its ResizeObserver, and a remounted row re-measures. ToolRow diffs files while present, so a closing edit keeps its body.
• --cv-fold-duration resolves on :root from --agent-motion-move, which applyAgentTheme sets on documentElement (composerStates.css default 190ms). The chevron transition uses the same token, only under no-preference.
• No literal colors in the diff. No new user-facing strings (tools.more only moved; the gallery note is dev-only). REVIEW.md: tests assert the new behavior and would fail without the fix. foldMotion as a test seam follows the existing thoughtLayout pattern.

CI (glaeda status, head 5f7be0d): the fold suites pass in CI: transcript-fold-motion, transcript-fold-anchor, transcript-fold-anchor-browser and ThoughtRow.test. web / react-apps-check shards 1–3 fail on the same tests on #23136, an unrelated revert on the same base: the typecheck exit from the warning in sidebar-rail-css.test.tsx, Ctrl-B, the SSH running turn, ↑/↓ prompt history, the @ picker, queue stack and composer strip. Shard 4 fails on composer-suggested-prompt and the Full access chip color. The diff doesn't touch either, but #23136's shard 4 was still running, so it isn't confirmed on base yet.

Not run: no local tsc or bun test (blocked on this host by policy), and no live app or gallery play of fold-holds-clicked-line.

[teamleaderleo]

Red checks on 5f7be0d are base reds, not this PR's: every failing test in web / react-apps-check (1)-(4) also fails on unrelated #23131 (run 39561886342), e.g. composer history, @-picker, Ctrl-B hint, suggested prompt, Full access chip colour, location strip, queue stack. The typecheck step exits 1 on 0 errors / 1 base warning. The fold tests this PR adds pass. Merging after review SHIP at this head.

[github-actions]

Shipped in:

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