[teamleaderleo]

HOLD: with adoption on, an adopted reopen is slower than a cold open, and under load adopted opens still time out (table below). Not to land until an adopted open beats the cold one idle and under load.

Leo, 2026-11-01: "i had to retry to get it to load???" and "i hate the "Starting Claude Code... cancel" thing that it starts with then the stupid ass layout shift surely we can have a better load/faster load/intro than that"

Opening an existing chat adopts the app's prewarmed agent page again (#27759, #31033), and the adopted page now shows its chat on the first try. #31486 had turned adoption off because of this bug. This turns it back on.

Cause

The parked spare page can be hidden: Chromium marks the page hidden, so document.visibilityState is hidden. A hidden document runs no animation frames.

The page told the host it drew its chat (pane.painted) one animation frame after that render. The host keeps an adopted page away and covered until that report (AgentPaneView+SessionAdopt.swift), and the cover keeps it hidden. So the report never came. Every reopen held its cover until the loading state gave up, about 15 s later, and so did some sidebar opens.

A diagnostic bench at 7d8c57e read this from the page itself in each failed open:

• visibilityState was "hidden";
• the last animation frame came before the report was asked for;
• the chat itself was drawn (88 transcript rows, chatRead, showsWhatItIs).

Fix

• paintReport.ts afterDrawnFrame. The report still waits one frame, then a task. A page that is hidden, or goes hidden while it waits, reports on a task instead. The host then shows the page, and its own leave already re-syncs Chromium's visibility (AgentPaneLoadingView.leave).
  • One visibilitychange listener, removed when the report runs (stop condition). It reads no data, and it is listed in check-switch-fetches.py.
• The chat-open trace. An adopted page counts as drawn once its pane painted (traceSurface). The page's own first frame (page_painted) came before the trace started, back when it drew New Tab, so every adopted open read page_empty in the bench.
• AgentPaneSpareReadiness.adoptsExistingChats is gone. That was cmux NEXT: existing chats open cold again until the adopted spare page draws its chat #31486's switch, so the spare serves existing chats again.

Takes back #31486: it switched adoption off until the adopted page drew its chat; this PR is that fix, so adoption is on again.

Changelog

Fixed: opening an existing chat in cmux NEXT draws it from the prewarmed page on the first try, without the long loading cover.

Tests

webviews/test/paint-report.test.ts:

• a hidden page reports although it draws no frame;
• a shown page reports after its frame;
• a page hidden while it waits reports exactly once.

Command: cd webviews && bun test test/paint-report.test.ts.

• Red on the first commit (the old behaviour, moved unchanged into paintReport.ts): 2 fail.
• Green at the head: 3 pass.

test/spare-session-adopt.test.ts passes at the head.

Performance

scripts/measure/chat-open.sh (chat-open-bench.py: milliseconds from the action to the first frame of chat content, median of 5), run on the same fleet display Mac (Mac16,10) for both columns. Before is the feat-cmux-next tip this head merges (a0741bc, adoption off since #31486: cold opens); after is this head.

metric	before (a0741bc, cold)	after (fe66961, adopted)	delta	harness or run link
sidebar open, idle (median ms)	770.6	780.2	+9.6	chat-open.sh, 5 runs
sidebar open, load	2508.9	2 of 5 valid	regressed	same
reopen, idle	806.2	1115.3	+309	same
reopen, load	2476.5	2 of 5 valid (13.6 s, 15.0 s)	regressed	same
flashes, idle sidebar / reopen	0 / 0	3 / 5	worse	same

At e143c04 (the paint fix only), idle sidebar was 628.6 ms and idle reopen 1092.5 ms. No idle open hangs any more: the ~15 s cover is gone. But an adopted page takes about 800 ms from adoption to its paint report, where a cold page takes about 600 ms in total, and under load adopted opens still wait out the page's 10 s hold. The likely cause is under investigation: the parked spare was seen hidden before its adoption (diagnostic run), and Chromium throttles a hidden page's tasks.

Timers and observers

One visibilitychange listener per paint report, removed when the report runs. No new timers.

Observed / Not run

• Ran: bun run check in webviews (green), the cmux-next pre-push gate, both bun suites above, and the diagnostic bench described above.
• Ran: the CmuxNext compile check at e143c04 (passed).
• Pending: the before/after bench.

Risk: normal. This turns the warm-spare open path back on for every existing local chat. A spare that is not connected still opens the chat cold (servesChat).
Agent session: agent session

🤖 Generated with Claude Code

[github-actions]

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

Review fe66961: 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=pr31619 -f ref=lane/warm-spare-adopt-fix-b -f platform=macos -f backend_mode=local

Gallery before/after for b9bc608: 16 states rendered, 2 images differ, 2993 more touched states not rendered. Images are in the run's artifact, not on this public PR; the independent review of this head reads them on its own, so the PR needs no hand-made screenshots of these states.

gh run download 41480478226 -R manaflow-ai/cmux -n gallery-advisory-31619-b9bc6086f36f1c1b97e0f1f4f5ecd6ae252a6855 (then read gallery-review.json)

States

State	Theme	Status	Diff	Picked by
agent-pane.chat-width/standard	Apple System Colors Light	unchanged	0%	own App.tsx
agent-pane.chat-width/standard	Apple System Colors	unchanged	0%	own App.tsx
agent-pane.chat-width/wide	Apple System Colors Light	unchanged	0%	own App.tsx
agent-pane.chat-width/wide	Apple System Colors	unchanged	0%	own App.tsx
agent-pane.chat-width/wide-blocks	Apple System Colors Light	unchanged	0%	own App.tsx
agent-pane.chat-width/wide-blocks	Apple System Colors	unchanged	0%	own App.tsx
agent-pane.composer-edge/short-edge	Apple System Colors	changed	0%	own App.tsx
agent-pane.composer-edge/short-edge	Apple System Colors Light	unchanged	0%	own App.tsx
agent-pane.composer-edge/wide-edge	Apple System Colors	changed	0%	own App.tsx
agent-pane.composer-edge/wide-edge	Apple System Colors Light	unchanged	0%	own App.tsx
agent-pane.composer-edge/no-edge	Apple System Colors Light	unchanged	0%	own App.tsx
agent-pane.composer-edge/no-edge	Apple System Colors	unchanged	0%	own App.tsx
agent-pane.composer-floats/scrolled-to-end	Apple System Colors Light	unchanged	0%	own App.tsx
agent-pane.composer-floats/scrolled-to-end	Apple System Colors	unchanged	0%	own App.tsx
agent-pane.composer-floats/composer-grown	Apple System Colors Light	unchanged	0%	own App.tsx
agent-pane.composer-floats/composer-grown	Apple System Colors	unchanged	0%	own App.tsx
agent-pane.composer-rest-gap/gap-40	Apple System Colors Light	unchanged	0%	own App.tsx
agent-pane.composer-rest-gap/gap-40	Apple System Colors	unchanged	0%	own App.tsx
agent-pane.composer-rest-gap/gap-56	Apple System Colors Light	unchanged	0%	own App.tsx
agent-pane.composer-rest-gap/gap-56	Apple System Colors	unchanged	0%	own App.tsx
agent-pane.composer-rest-gap/gap-72	Apple System Colors Light	unchanged	0%	own App.tsx
agent-pane.composer-rest-gap/gap-72	Apple System Colors	unchanged	0%	own App.tsx
agent-pane.continuation-spacing/shipped	Apple System Colors Light	unchanged	0%	own App.tsx
agent-pane.continuation-spacing/shipped	Apple System Colors	unchanged	0%	own App.tsx
agent-pane.continuation-spacing/two-thirds	Apple System Colors Light	unchanged	0%	own App.tsx
agent-pane.continuation-spacing/two-thirds	Apple System Colors	unchanged	0%	own App.tsx
agent-pane.continuation-spacing/half-tucked	Apple System Colors Light	unchanged	0%	own App.tsx
agent-pane.continuation-spacing/half-tucked	Apple System Colors	unchanged	0%	own App.tsx
agent-pane.continuation-spacing/before	Apple System Colors Light	unchanged	0%	own App.tsx
agent-pane.continuation-spacing/before	Apple System Colors	unchanged	0%	own App.tsx
agent-pane.copy-controls/output	Apple System Colors Light	unchanged	0%	own App.tsx
agent-pane.copy-controls/output	Apple System Colors	unchanged	0%	own App.tsx

_{Pick other states with a Gallery: entry.id, entry.id/variant line in the PR body.}

web and size checks: green

[teamleaderleo]

Review (claude-opus-5-5): SHIP at e143c04

Review (subagent claude-opus-5-5 medium; carried): SHIP at e143c04

The SHIP at 2d763e3 carries: the PR's own diff has the same git patch-id --verbatim (2f5384580e97) at both heads, and the merge of feat-cmux-next between them (for the Swift 6.3 crash fix #31525) resolved nothing by hand. CmuxNext compile check passed at this head.

Follow-ups: #31622.

[teamleaderleo]

Review (claude-opus-5-5): SHIP at fe66961

Review (subagent claude-opus-5-5 medium): SHIP at fe66961

Delta from e143c04 (SHIP above). adoptSession re-reads the list only after select() throws. Only a list that was read and lacks the chat marks it missing (stillOpen, or leftByRead when the read itself fell back from the chat). A stale fallback cannot match, because generations only increase. A user pick during the failed open is never overridden. The attach reply carries the title and summary, so an unlisted chat draws no worse, and the row fills in later through the watch and readLaterSessions. The new test is red on edebfcc and green here.

Follow-ups:

• [follow-up] Confirm that an unlisted chat on another machine, opened by adoption while that machine is down, still keeps the Mac's copy.
• [follow-up] Tests: an attach that fails while the list has the chat (says failed, not missing); a list read that fails after a failed attach; a pick during a failed adoption.
• [follow-up] Skip the read when a newer pick already replaced the adoption (direct.ts:2080).

[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.

---

For awareness: #32448 carries this PR's commits and adds the root cause its bench pointed at. The pane's paint hold made the incoming pane transparent, which hid the held-away adopted page from Chromium (#28002's rule), so the page drew only at the hold's 500 ms limit. #32448 keeps a page laid out outside the window visible to Chromium, and it does not carry the hidden-page task report (a likely source of the flashes). Before/after chat-open bench runs are queued on the fleet.