test(studio): keep the reasoning and activity roster pulses as status, not Skeletons

[peppescg]

Summary

#2337 asks for the Skeleton primitive (components/ui/skeleton.tsx) wherever animate-pulse marks content that is still loading. #2358 reviewed three call sites. This PR reviews the last two, in reasoning-disclosure.tsx and delegation-panel.tsx. Neither is a loading placeholder. Both pulse while something runs, so there are no code changes. The PR records the decision and pins it, which finishes the issue.

Closes #2337
Part of #2201
Part of #2210

Decisions

File	Occurrence	What it marks	Decision
features/chat/reasoning-disclosure.tsx:46	shrink-0 animate-pulse font-medium on the "Reasoning…" label	The model's reasoning is streaming. The label is real text, and it sits beside the latest reasoning line as a live preview. Once the turn settles, the label is replaced by the "Reasoning summary · N lines" toggle.	Status: keep.
features/chat/delegation-panel.tsx:216	RosterGlyph: size-2 animate-pulse rounded-full bg-brand	A roster row in Session activity is running. When the row settles, the dot becomes a check, an alert, or a help icon.	Status: keep.

The roster dot is decorative (aria-hidden), and the row's text names the state ("Parallel group d2 · Running"). The reasoning label isn't hidden, because it's the spoken text itself. Neither one stands in for content that hasn't arrived yet: the preview line beside the label and the row text are the content. The prototype (stack-08) pulses the same "Reasoning…" label and the same brand dot on running agent rows.

Why delegation-panel isn't stacked on #2334. delegation-panel.tsx is also changed in #2334. That PR moves the roster rows onto the Toggle primitive, but it leaves RosterGlyph alone. Because the decision is "keep" and this PR changes only the new test file, it doesn't conflict with #2334. The new test passes on main (8cc7c2347) and on #2334's head (acfbf3382), so I folded it in here and didn't open a stacked PR. It selects roster rows by button[aria-pressed], which matches both the current <button> and #2334's Toggle.

What changed

features/chat/status-pulse.pinning.test.tsx (from #2358) gets 2 new tests, written against the unchanged code. The file now has 6 tests, and its header comment covers the two new call sites:

• Activity roster: a running row pulses its dot. The dot is aria-hidden, round, bg-brand, has no text, and isn't a Skeleton. A finished row in the same panel doesn't pulse, so the panel has exactly one pulse, and an all-finished roster has none.
• Reasoning: while reasoning streams, there is exactly one pulse. It's the "Reasoning…" label: not aria-hidden, not a Skeleton, and followed by the preview line. The settled summary has no pulse, and empty reasoning renders nothing.

No production file changes.

No regression

• Mutation proof. I applied each mutation alone, reran the pin file, and reverted it. Each one fails exactly one test:

  Mutation	Failing test
  Roster running dot becomes a Skeleton (data-slot="skeleton")	activity roster
  Roster running dot loses aria-hidden	activity roster
  Roster finished glyph (Check) pulses too	activity roster
  "Reasoning…" label replaced by a Skeleton bar	reasoning
  "Reasoning…" label becomes aria-hidden	reasoning
  Settled "Reasoning summary" toggle pulses	reasoning
  Reasoning pulse removed	reasoning

• Cross-branch check. The same pin file passes on refactor(studio): use the Tabs and Toggle primitives in two hand-rolled controls #2334's head (acfbf3382), with 6 of 6 tests.

• Gates (through the shared lock):

  • task studio:check: lint, typecheck and generated-check pass. Web: 188 files, 1618 of 1618 tests pass. Server: 27 files, 252 of 252 tests pass.
  • task studio:browser: 136 passed, 38 skipped.

Screenshots

Before is origin/main (8cc7c2347), After is this branch (e83afc10a), and Prototype is stack-08. Every capture is pixel-identical to Before. A second Before capture also matched (noise 0); pixel-report.txt in the asset folder has the counts. There are no red boxes, because nothing changed. Amber boxes mark the reviewed status pulses instead. To find them, a separate capture pass hid only those pulses and diffed the result.

Reasoning running, desktop, light. The box is the "Reasoning…" label. After matches the prototype: both show the same pulsing label and the same one-line preview. The other differences are outside this PR: the indicator under the turn reads "Thinking" here and "Reasoning" in the prototype, and the layout and composer copy differ. The prototype's harness streamed the same reasoning text.

Running activity in the Session activity panel, desktop, light. The box is the running parallel group's roster dot. After matches the prototype's running-row dot. The prototype's Agents panel opens on Subagents and lists branches as rows, where Studio lists the parallel group. That's a layout difference this PR doesn't cover.

Desktop dark and 390px, light and dark

Reasoning running, desktop, dark:

Reasoning running, 390px, light:

Reasoning running, 390px, dark:

Session activity, desktop, dark:

Session activity, 390px, light:

Session activity, 390px, dark:

🤖 Generated with Claude Code

[kantord]

Thanks for finishing the animate-pulse audit, and for checking the roster test against #2334's Toggle rows (button[aria-pressed] was a good choice of selector). Approving, with one non-blocking nit:

• Same as on #2358: this pins a "keep it a status dot" decision with tests, and some assertions pin incidental details (font-medium, nextElementSibling) rather than the decision. The data-slot="skeleton" absence check and the textContent check are the ones that would actually catch a Skeleton swap. A code comment next to the decision might carry it more cheaply than more class-level assertions.

Fully or partially written by an AI agent.