test(studio): keep the chat status pulses as status dots, not Skeletons

[peppescg]

Summary

#2337 asks for the Skeleton primitive (components/ui/skeleton.tsx) wherever animate-pulse marks content that is still loading. This PR reviews the three call sites that no other open PR touches. None of them is a loading placeholder. Each is a status dot that pulses while something runs, so no code changes. The PR records that decision and adds tests that pin it, so a later sweep can't turn these dots into Skeletons by mistake.

Part of #2337
Part of #2201
Part of #2210

Decisions

File	Occurrence	What it marks	Decision
features/chat/delegation-card.tsx:54	StateGlyph: size-2 animate-pulse rounded-full bg-brand	A subagent, parallel group, or branch is running. Once it settles, the dot becomes an alert icon or a branch icon.	Status: keep.
features/chat/fleet-status-chip.tsx:132	size-1.5 animate-pulse rounded-full bg-brand in a chip segment	That family (subagents, parallel, teams) has running work (segment.running).	Status: keep.
features/chat/session-details-dialog.tsx:43	STATE_DOT.running: bg-brand animate-pulse	The session's lifecycle state is running. The other states use static coloured dots.	Status: keep.

Each dot is decorative (aria-hidden). The state is spoken as text beside it ("Running", the segment's aria-label, or "running"), and the dot disappears once the work settles. The prototype (stack-08) shows the same pulsing dots on all three surfaces; see the Prototype column below.

The details view's real loading state ("Loading details…") is a spinner (Loader2 animate-spin) in a role="status" paragraph, not an animate-pulse placeholder, so #2337 doesn't cover it. It stays unchanged, and is pinned too.

Later: these files from the issue are in open PRs. Each one gets its decision once that PR lands:

File	Open PR
features/chat/chat-composer.tsx	#2331
features/chat/delegation-panel.tsx	#2334
features/chat/reasoning-disclosure.tsx	#2352
features/chat/tool-call-list.tsx	#2321

What changed

• features/chat/status-pulse.pinning.test.tsx (new, 4 tests), against the unchanged code:

  • a delegation card pulses its dot only while the child runs; the dot is aria-hidden, and "Running" sits in the card's live region;
  • a fleet chip segment pulses only while its family has running work (data-running), and keeps its accessible name (Subagents · 1 running · 1 done) inside the Agents in this chat list;
  • the session details state dot pulses for running only; idle, completed, failed, awaiting and cancelled don't pulse;
  • the details loading state is a role="status" with "Loading details…" and an aria-hidden spinner; it has no pulse, no Skeleton, and no detail list.

  Every pulse check also asserts that the dot isn't the Skeleton primitive ([data-slot="skeleton"]).

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
  Delegation card dot becomes a Skeleton (data-slot="skeleton")	delegation card
  Delegation card dot loses aria-hidden	delegation card
  Fleet chip pulse removed	fleet chip
  Session details completed dot pulses too	session details state dot
  Details loading spinner becomes animate-pulse	details loading state

• Gates (through the shared lock):

  • task studio:check:
    • lint, typecheck and generated-check pass;
    • web: 176 files and 1533/1533 tests pass;
    • server: 240/241. The one failure is knowledge.test.ts › "declares the transport errors…" (expected 15, got 16). It already fails on main (4aa238f3c, the CI run for that commit) because feat(studio): show a skill's files on its page #2155 and fix(studio): declare 400 and 429 on the knowledge routes #2313 collided. fix(studio): declare 400 and 429 on the skill files route, and stop hard-coding the route count #2355 has since fixed it on main, and the CI Studio job for this PR passes. This PR doesn't touch apps/server.
  • task studio:browser: 132 passed, 38 skipped.

Screenshots

Before is origin/main (4aa238f3c), After is this branch (d49cc4714), and Prototype is stack-08. Every capture is pixel-identical to Before, and 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 dots instead. A separate capture pass hid only those dots and diffed the result to find them.

After matches the prototype: the prototype shows the same pulsing brand dots for running delegations, the fleet chip, and the running session state. The other differences in the Prototype column are layout and copy (for example, "Running" capitalised in its details dialog). This PR doesn't cover them.

Running delegation and fleet chip, desktop, light. The boxes are the fleet chip's Parallel segment dot and the dots of the running parallel group and branches 2 and 3. Branches 2 and 3 merge into one box.

Session details for a running session, desktop, light. The box beside "running" is the state dot. The dimmed chat behind the dialog shows the same delegation dots.

Desktop dark and 390px, light and dark

Running delegation, desktop, dark:

Running delegation, 390px, light. The chat hides the fleet chip in a narrow header, so only the card dots are boxed:

Running delegation, 390px, dark:

Session details, desktop, dark:

Session details, 390px, light:

Session details, 390px, dark:

🤖 Generated with Claude Code

[kantord]

Thanks for this, recording why these three dots stay dots is useful, and the decision table is clear. Approving, with a few non-blocking nits:

• This is 188 lines of tests for no code change, so it is worth asking whether a code comment or the decision table alone would carry the decision more cheaply.
• Only the [data-slot="skeleton"] absence check guards against a Skeleton swap. The .animate-pulse count would still pass, since Skeleton also pulses, and the rounded-full and bg-brand checks pin styling rather than the decision.
• It covers 3 of the animate-pulse sites, so other pulse sites remain unpinned.

Fully or partially written by an AI agent.