[teamleaderleo]

Rebased on #29608 and #29264. Leo's later words in #29608 ("the leading icons — just the stupid fork one", "why didn't you guys remove the icons nor the tokens") supersede two parts of this PR. A subagent line now leads with its harness mark or toned icon (the spinner while it runs) and never the fork. It shows no tokens or tool uses: the trailing meta holds only a workflow's agent counts and a state worth a word. Stop and the chevron keep #29264's tracks. The stills below predate both, so they still show the fork and the token meta. What this PR still adds: no agent-type subtitle (it moves to the tooltip), no Done or Running word (screen readers still get it), the meta before the time, and no card when nothing runs. Follow-up from review: #29616.

Leo's words

On the latest NEXT nightly: "why is this still so ass in latest nightly". Earlier: "eg a lot of these subtitles are bad".

The finished subagent line he showed had two leading glyphs (a chat bubble, then the fork, the fork touching the title), a "general-purpose" subtitle, its duration and "Done" in two columns far right, inside a bordered card for one finished line. The running line had the spinner and the fork, a "Bash" subtitle, and its tokens and tool uses on a second line.

What changed

Every subagent, started-chat and command line in the running tray (subagents/AgentLine.tsx) now has one anatomy, after the subagent and task rows in Claude desktop and Claude Code:

• One leading glyph. The spinner while it works, otherwise the fork (subagent), the chat (started chat) or the terminal (command), or the harness mark. The second kind glyph (carry of #27602: compact waiting messages and subagent tray grid #28113) is gone, and so is the glyph touching the title: the title sits the row's gap after the one glyph.
• No agent-type subtitle. The kind (Explore, general-purpose) and what it does now move to the line's tooltip (Inspector: its hover card). Nothing is deleted.
• One muted trailing meta, then the time, then the chevron. Tokens and tool uses sit muted right before the time, lined up with finished lines' times. Done and Running are no longer words on the line, since the glyph says them; screen readers still read them. Waiting, Failed, Stopped and Retrying keep their word in the meta. The 96px status column shrinks to the chevron.
• No card when nothing runs. A tray holding only finished lines draws no fill or ring (data-settled). The card comes back as soon as something runs.
• Sibling rows that broke the rule: the Inspector's and Session Info's Subagents lists lose the kind subtitle (it moves to the hover card or tooltip), and the subagent group's open list (cv-subagent__kind) does too.

Unchanged: what a line opens and its content, Stop, the running spinner and wave, the narrow tray's two-line layout (stats wrap under the title there), and the transcript's "Subagent finished" line. That line already has one glyph and no subtitle, and its trailing column belongs to #28860, which this PR doesn't touch.

Gallery: 3 variants

agent-pane.subagent-row-anatomy with experiment subagent-tray-row, across the done-one, running-stats and mixed fixtures:

• trailing (shipped): stats muted before the time, on one line.
• below: stats on a second line under the title (closest to the earlier look).
• inline: stats and time right after the title, with no trailing column.

I picked trailing. It keeps every line to one row, lines the running line's time up with the finished lines' times, and matches the transcript's one row grammar (status, then duration, then chevron). inline reads well for one line but loses the shared time column across several. below doubles the row height for two numbers.

Before / after

Linux gallery (Chromium, 2x), synthetic fixtures agent-pane.subagent-row-anatomy, captured by subagent-tray-columns.test.ts; before is feat-cmux-next 7ad4616221fa's components on the same fixtures.

One finished subagent, nothing running

	Before	After
Light
Dark

Running subagent with tokens and tool uses, over a finished one

	Before	After
Light
Dark

Running, failed and finished (an Explore agent)

	Before	After
Light
Dark

I viewed the after images.

Tests

• subagent-tray-columns.test.ts "row anatomy" (real Chromium, light and dark, three fixtures): one leading glyph; title at least --ds-space-2 after it; no general-purpose/Explore/Done/Running visible on the line; the meta ends before the time; a settled tray has no shadow and no fill; a long state word is cut inside its meta.
• subagent-tray.test.tsx: one glyph (spinner while running, fork once ended), the name first in the text, the kind and the action in the tooltip, stats kept on the line, the screen-reader state word, settled vs not.
• subagent-progress.test.tsx: the group list's rows no longer carry the kind.

cmux-ci (Blacksmith Linux, fast-checks) at 97a37b9 (the head minus a format-only commit): subagent-tray.test.tsx, subagent-progress.test.tsx, running-wave, session-info-subagent-marks, chat-inspector 77 pass 0 fail; subagent-tray-columns.test.ts 16 pass 0 fail (the existing narrow/short-dock/columns geometry plus the six new anatomy tests); bun run check green (design tokens, fmt, lint, types). run 41207645140.

Spec

leo-workspace design/chat-transcript.md "Subagent row anatomy" (teamleaderleo/leo-workspace#1912).

Performance

metric	before	after	delta	harness or run link
per-token work in tray lines	none	none	0	no new timers, observers, effects or state; one useState read of the experiment arm per tray mount, as the panel and motion arms already do

Risk: low to medium. Markup and CSS for the tray's lines and the Subagents lists. The narrow (two-line) tray's grid gains a column for the meta. Its geometry tests (subagent-tray-columns.test.ts) still run.

Changelog

Changed: running and finished subagent lines show one icon, the title, muted stats and time; the agent type moves to the tooltip, and a tray with nothing running draws no card.

Agent session: agent session

🤖 Generated with Claude Code

[blacksmith-sh]

Found 3 test failures on Blacksmith runners:

Failures

Test	View Logs
tab: tabGroup.addTab is not a top-level row/tab: tabGroup.addTab is not a top-level row	View Logs
tab: tabGroup.create is not a top-level row/tab: tabGroup.create is not a top-level row	View Logs
tab: tabGroup.removeTab is not a top-level row/ tab: tabGroup.removeTab is not a top-level row	View Logs

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

[github-actions]

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

Review 896d0c7: 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=pr29276 -f ref=lane/subagent-row-anatomy -f platform=macos -f backend_mode=local

Gallery before/after for 5bf6cd8: 16 states rendered, 10 images differ, 1182 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 41246553092 -R manaflow-ai/cmux -n gallery-advisory-29276-5bf6cd83efe87f6e261e92a1978303ca905687ba (then read gallery-review.json)

States

State	Theme	Status	Diff	Picked by
agent-pane.background-activity/duration-columns	Apple System Colors Light	unchanged	0%	own subagents.css
agent-pane.background-activity/duration-columns	Apple System Colors	unchanged	0%	own subagents.css
agent-pane.background-activity/pinned-live-tail-light	Apple System Colors Light	unchanged	0%	own subagents.css
agent-pane.background-activity/pinned-live-tail-light	Apple System Colors	unchanged	0%	own subagents.css
agent-pane.background-activity/pinned-live-tail-dark	Apple System Colors Light	unchanged	0%	own subagents.css
agent-pane.background-activity/pinned-live-tail-dark	Apple System Colors	unchanged	0%	own subagents.css
agent-pane.background-activity/after-latest-message-light	Apple System Colors Light	unchanged	0%	own subagents.css
agent-pane.background-activity/after-latest-message-light	Apple System Colors	unchanged	0%	own subagents.css
agent-pane.background-activity/after-latest-message-dark	Apple System Colors Light	unchanged	0%	own subagents.css
agent-pane.background-activity/after-latest-message-dark	Apple System Colors	unchanged	0%	own subagents.css
agent-pane.background-activity/compact-pill-light	Apple System Colors Light	unchanged	0%	own subagents.css
agent-pane.background-activity/compact-pill-light	Apple System Colors	unchanged	0%	own subagents.css
agent-pane.background-activity/compact-pill-dark	Apple System Colors Light	unchanged	0%	own subagents.css
agent-pane.background-activity/compact-pill-dark	Apple System Colors	unchanged	0%	own subagents.css
agent-pane.background-activity/side-panel-light	Apple System Colors Light	unchanged	0%	own subagents.css
agent-pane.background-activity/side-panel-light	Apple System Colors	unchanged	0%	own subagents.css
agent-pane.chat-inspector/full	Apple System Colors Light	unchanged	0%	own SubagentsSection.tsx
agent-pane.chat-inspector/full	Apple System Colors	unchanged	0%	own SubagentsSection.tsx
agent-pane.chat-inspector/workflow-running	Apple System Colors	changed	0.0043%	own SubagentsSection.tsx
agent-pane.chat-inspector/workflow-running	Apple System Colors Light	changed	0.0042%	own SubagentsSection.tsx
agent-pane.chat-inspector/workflow-finished	Apple System Colors Light	unchanged	0%	own SubagentsSection.tsx
agent-pane.chat-inspector/workflow-finished	Apple System Colors	unchanged	0%	own SubagentsSection.tsx
agent-pane.chat-inspector/workflow-agent-row-compact	Apple System Colors Light	changed	0.0043%	own SubagentsSection.tsx
agent-pane.chat-inspector/workflow-agent-row-compact	Apple System Colors	changed	0.0038%	own SubagentsSection.tsx
agent-pane.chat-inspector/workflow-agent-row-detail	Apple System Colors Light	changed	0.0044%	own SubagentsSection.tsx
agent-pane.chat-inspector/workflow-agent-row-detail	Apple System Colors	changed	0.0036%	own SubagentsSection.tsx
agent-pane.chat-inspector/workflow-agent-row-grid	Apple System Colors Light	changed	0.005%	own SubagentsSection.tsx
agent-pane.chat-inspector/workflow-agent-row-grid	Apple System Colors	changed	0.0041%	own SubagentsSection.tsx
agent-pane.chat-inspector/workflow-agent-row-narrow	Apple System Colors Light	changed	0.0051%	own SubagentsSection.tsx
agent-pane.chat-inspector/workflow-agent-row-narrow	Apple System Colors	changed	0.0039%	own SubagentsSection.tsx
agent-pane.chat-inspector/reorder-dragging	Apple System Colors Light	unchanged	0%	own SubagentsSection.tsx
agent-pane.chat-inspector/reorder-dragging	Apple System Colors	unchanged	0%	own SubagentsSection.tsx

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

Passes: CI passes on 5bf6cd83ef.

CI passes on 5bf6cd83ef (run 41246553374 attempt 1).

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.

web and size checks: green

Shipped in:

• cmux NEXT nightly 1.0.0-nightly.4125243132903 (b06e339a1b)
• cmux NEXT nightly 1.0.0-nightly.4125324323503 (35bf9b2ff4)
• cmux NEXT nightly 1.0.0-nightly.4125408606503 (3f2c154c09)
• cmux NEXT nightly 1.0.0-nightly.4125648764703 (eb3bf75592)

[teamleaderleo]

Review (codex gpt-6-astra high, next-review): SHIP at 000d80f

SHIP at 000d80f

Visual verdict: supplied after screenshots look correct; no repeated controls or off-screen layout.

1. [P2][follow-up] The hidden state text is not exposed to screen readers. Tray buttons set aria-label at AgentLine.tsx:136, which overrides descendant content, so the .acpmux-agent-line__sr “Running”/“Done” text at line 179 is ignored. Include the state (and relevant time/details) in the accessible label or remove the overriding label.

Follow-ups:

• [P2][follow-up] Preserve the quiet row’s state in the button’s accessible name.

[cmux-bot]

cmux-bot approves 000d80f: an independent review passed this exact head (next-review receipt by teamleaderleo). A push needs a new review; a later BLOCK at this head dismisses this approval.

[teamleaderleo]

Review (codex, router-review): SHIP at 000d80f

Reviewer: codex gpt-6-astra high, fast tier.

Follow-ups: 2. Each lands as an issue in this repo that links this receipt; the lane that opened the PR fixes it forward.

SHIP at 000d80f

Visual verdict: no images; Leo tries the build

1. [P2] [follow-up] The new screen-reader-only state text is suppressed by the button’s explicit aria-label: AgentLine.tsx:136 overrides descendant content, so the Running/Done span added at AgentLine.tsx:179 is not announced for tray subagent rows. Include the state in the accessible label or reference it with aria-describedby.

Follow-ups:

• [P2] [follow-up] Fix the quiet-state accessible name as described above.
• [P2] [follow-up] No after screenshots were supplied for visual verification.

[teamleaderleo]

Review (codex gpt-6-astra high, next-review): SHIP at 2f1f343

SHIP at 2f1f343

Visual verdict: supplied light/dark screenshots reviewed; no layout or control regressions found.

Follow-ups: none

[cmux-bot]

cmux-bot approves 2f1f343: an independent review passed this exact head (next-review receipt by teamleaderleo). A push needs a new review; a later BLOCK at this head dismisses this approval.

[teamleaderleo]

Heads-up from #29608: Leo, 2026-10-31, on the subagent row: "why didn't you guys remove the icons nor the tokens", then "the leading icons — just the stupid fork one". #29608 removes the fork glyph (kindIcon agent.fork) from tray lines and drops tokens and tool uses from the tray and transcript group rows (subagentProgress(..., { usage: false })). This PR makes the fork the one glyph and keeps tokens as trailing meta, so when you rebase, please keep the harness mark, drop the fork, and drop the tokens and tool uses.

[teamleaderleo]

Review (codex, router-review): SHIP at 51d82ee

Reviewer: codex gpt-6-astra high, fast tier.

Follow-ups: 1. Each lands as an issue in this repo that links this receipt; the lane that opened the PR fixes it forward.

SHIP at 51d82ee

Visual verdict: Reviewed all 22 after images against their before renders. The compact, Claude-style tray rows look clearer; no visible layout regression in the supplied states.

1. [P2] [follow-up] subagents.css:508: Stats take width before the title. The title retains flex: 1 (zero basis), while meta reserves its content width up to 50%. Just above the 400px narrow-layout cutoff, long stats leave only a few characters for the title; indentation can exhaust that space. Reserve usable title width and truncate stats first. Add coverage around 401–450px with long stats and nested agents.

All 13 changed TypeScript files parsed successfully. Tests could not run locally because Bun and webview dependencies are unavailable.

Follow-ups:

• Preserve readable titles beside long stats near the narrow-layout breakpoint.

[teamleaderleo]

Review (codex, router-review): SHIP at 896d0c7

Reviewer: codex gpt-6-astra high, fast tier.

Follow-ups: 1. Each lands as an issue in this repo that links this receipt; the lane that opened the PR fixes it forward.

SHIP at 896d0c7

Visual verdict: Reviewed all 22 after images against their before pairs. No new visible overflow or duplicated controls. PR-body tray screenshots predate the rebase.

1. [P2] [follow-up] subagents.css:525: The separator breaks the single-line layout when workflow counts accompany Waiting, Failed, or Retrying. The state uses Tick, an inline grid whose text occupies cell 1 / 1; its new ::before becomes an automatically placed item in a second row. Move the separator outside Tick and cover a workflow with both counts and a visible status.

Tests could not run locally: Bun and dependencies are absent.

Follow-ups:

• Fix the separator’s grid placement and test counts alongside a visible status.

[teamleaderleo]

Review (codex, router-review): SHIP at 5bf6cd8

Reviewer: codex gpt-6-astra high, fast tier.

SHIP at 5bf6cd8

Visual verdict: Reviewed all 22 after images against their before images. No blocking visual regression; the simplified rows fit the compact Claude-style layout. PR-body stills predate the rebase.

No correctness findings. Tests weren’t rerun: Bun and dependencies are unavailable in this box.

Follow-ups: none