Review — end-to-end ✅ (clean approve)

What this ships

UX polish on top of #1203's profile-switch correctness fix. Two changes:

1. Spinner + optimistic chip label (panels.js:2014-2023 + style.css:661-663) — On click: .switching CSS class + _chip.disabled = true + immediate _chipLabel.textContent = name. The user sees the new name in <1ms; spinner ring rotates on the icon while fetches complete; finally block always cleans up; on error the label reverts to the previous name.

2. Parallelized fetches (panels.js:2033-2037) — populateModelDropdown() and loadWorkspaceList() are independent network calls (one hits /api/models, the other /api/workspaces) that were running sequentially. Now they run via Promise.all. Apply steps stay after the join — the order-sensitive logic (S._pendingProfileModel, workspace update, session update) is unchanged.

Traced against upstream hermes-agent

Pure WebUI presentation. Both endpoints (/api/models and /api/workspaces) are independent in the agent's view too — there's no inter-endpoint state ordering to preserve. ✅

End-to-end trace

Spinner state machine

Error path:

The state machine is sound — the finally block guarantees cleanup, and both disabled=true (HTML attribute) + pointer-events:none (CSS) prevent stacked clicks during the switch.

Promise.all parallelization

populateModelDropdown() is async at ui.js:133; loadWorkspaceList() is async at panels.js:1299. Neither reads the other's output. Switching from sequential awaits to Promise.all([...]) is safe — the slower of the two bounds the wall time. The PR description's "~50% faster" claim matches this analysis when the two fetches have similar latency.

The post-join apply steps (_pendingProfileModel, workspace input update, session update) all run in the same code path as before, just unindented from the previous sequential await chain. Order-sensitive correctness preserved.

@keyframes spin reuse

The new .composer-profile-chip.switching .composer-profile-icon::after rule references animation:spin .6s linear infinite. @keyframes spin is defined at style.css:323 (re-used by other spinner CSS at lines 271, 1643). ✅ No new keyframe needed.

Spinner positioning

.composer-profile-chip.switching .composer-profile-icon::after {
  content:''; display:inline-block; width:10px; height:10px;
  border:1.5px solid currentColor; border-top-color:transparent;
  border-radius:50%;
  animation:spin .6s linear infinite;
  position:absolute; left:50%; top:50%;
  margin-left:-5px; margin-top:-5px;
}
.composer-profile-chip.switching .composer-profile-icon { position:relative; }

The icon container becomes position:relative only when .switching is applied, so the absolute-positioned ::after ring is anchored to the icon. Outside switching state, the icon's positioning is unchanged. ✅

Edge-case trace

Tests

• PR's 11 new tests (tests/test_profile_switch_ux.py): all pass.
  • Spinner CSS class added + removed in finally
  • Optimistic name update before API call
  • Chip disabled + re-enabled in finally
  • Error path reverts label to _prevProfileName
  • Promise.all pattern present + sequential pattern absent
  • Apply steps come AFTER Promise.all resolves
  • CSS .switching class has cursor:wait + pointer-events:none
• Local full suite: 2753 passed, 47 skipped, 0 failures.
• CI on PR: ✅ test (3.11), ✅ test (3.12), ✅ test (3.13).

Other audit — confirmed correct

• JS syntax: node --check passes on panels.js.
• No agent coupling: pure UI state machine + frontend network parallelization.
• Builds on #1203 correctness: this PR adds UX polish; the underlying bugs (#1200 thread-local workspace, model cache, chip label) were fixed in #1203 and remain unaffected.
• finally is the right primitive: guarantees cleanup on early return, throw, or normal exit.
• _chip.disabled + CSS pointer-events:none is belt-and-suspenders against double-clicks. Either alone would suffice on modern browsers; together is defensive.

Minor observations (non-blocking)

• Catch reverts label even when POST succeeded: if /api/profile/switch returns 200 (so S.activeProfile = B is set) but a downstream step (populate/loadWorkspace/applyBotName) throws, the catch reverts the chip label to "A". The S.activeProfile is still "B" though — so next syncTopbar() call re-renders the label to "B". The UX wart is fleeting (chip says "A" briefly then snaps to "B"). Acceptable; the more correct version would be to track which step failed and only revert if the POST itself failed. Out of scope for this UX PR.
• The CHANGELOG entry at the very top of the file inserts before ## v0.50.225, but the previous batch's v0.50.230 heading was right above. Looking at the file: ### Bug fixes block at top is for v0.50.231 (probably the macOS quirk), then v0.50.230, then v0.50.225. The new v0.50.235 entry inserted between v0.50.231 and v0.50.225 — that's the right spot chronologically (newest first). ✅
• The 0.6s spin period is slightly faster than the existing 1s spin used in the workspace dropdown spinner. Cosmetic — both are reasonable.

Recommendation

Approved. Tight UX polish on top of #1203's correctness fixes. State machine is sound (finally guarantees cleanup, optimistic update reverts on error), parallelization is safe (independent fetches, apply steps post-join), and the spinner reuses the existing @keyframes spin. CI green; full suite clean. Parked at approval — ready for the release agent's merge/tag pipeline.