fix(desktop): keep Dark mode across Bot Mode new chat

[FalconOrtiz]

Why

Bot Mode New chat with this bot activates the bot gateway profile. That profile usually has no appearance of its own. When the launch profile is not default, Settings → Appearance Dark only wrote the per-profile record, so the hop resolved system and a light OS flipped the window to Light (#101216).

Scope

• apps/desktop/src/themes/context.tsx — profilePref.assign mirrors named-profile picks into the global slot; promoteUnanimousLegacy fills an empty global slot when every per-profile value agrees.
• apps/desktop/src/themes/profile-theme.test.ts — covers inherit, override, and promote for the Desktop: Bot Mode "New chat with this bot" (roster right-click) flips Dark mode to Light — per-profile theme follows a gateway activation the user didn't ask for #101216 path.

Out of scope: $appearanceProfile decoupling (issue suggestion 1), Theme Scope Shared (#96271), Appearance caption copy.

Tradeoffs

Mirroring into the global slot means the last named-profile Appearance pick becomes the fallback for never-themed profiles. Explicit per-profile overrides still win. This is smaller than teaching ThemeProvider to ignore Bot Mode gateway activations.

Blast Radius

Desktop theme/mode persistence only. Users with one unanimous mode across profiles get that mode on never-themed bots and on cold launch after a bot hop. Installs with conflicting per-profile modes keep those overrides and do not auto-promote.

Verification

cd apps/desktop
npx vitest run src/themes/profile-theme.test.ts src/themes/context.test.tsx

20 passed, exit 0.

[Enough1122]

AI code review — automated review for reference; please use your judgment.

Overall: Named-profile appearance assigns now mirror into the global slot (and a unanimous pre-existing record promotes into it), so a Bot Mode hop onto a never-themed bot inherits the look the user picked instead of resolving system (#101216).

Correctness:

• apps/desktop/src/themes/context.tsx:profilePref — default still writes only the legacy slot; named assigns write their record entry plus the legacy slot; explicit per-profile overrides still win on resolve (test-pinned).
• promoteUnanimousLegacy — no-ops when the legacy slot is set or records disagree; unanimous promotion covers pre-per-profile installs.

Non-blocking nit:

• resolve() now has a write side effect (promotion persists on read). Harmless here (idempotent, same value), but a reader-triggered write can surprise — e.g. under concurrent windows racing the first resolve; last-writer-wins with identical values, so no corruption, just noting it.

Verdict: LGTM.

[FalconOrtiz]

Requesting squash-merge per engineer agreement (2026-09-04): AI-review merge-ready pack; nits non-blocking; no rewrites.

This PR is in the agreed pack with #101523 #101528 #101529 #101530 #101531 #101537 #101544 #101547 #101548 #101549 #101551.

Out of pack (do not block this merge): #101564, #101796/#101091, #101318.

[kvnloo]

Exact-head review of 7335bd7.

Keep: named-profile assign mirrors into the global/legacy slot, plus inherit/override tests (#101216):

• never-themed bot inherits Dark set on the launch profile
• explicit per-profile override still wins
• unanimous pre-existing per-profile mode promotes into an empty global slot
  Without the mirror, Bot Mode hopping onto a never-themed bot falls through to system.

Check:

• last-assign fallback: every named assign writes persistString(legacy, value), so the global slot is whichever profile was assigned last. After forge=dark then atlas=light, a third never-themed profile inherits light. Confirm that is the intended hop, not "first launch profile."
• promoteUnanimousLegacy runs inside resolve (write-on-read). A getter that persists is surprising if two profiles already disagree it no-ops, but a single stored profile will stamp legacy on first resolve of any name. Worth a one-line note if that migration should be a one-shot boot path instead of every resolve.

No competing PR from me.

[FalconOrtiz]

@kvnloo Confirmed. Last named-profile assign is the intended hop, not the first launch profile.

assign writes persistString(legacy, value) on every named pick, so after forge=dark then atlas=light a third never-themed profile inherits light. Pack agreement 2026-09-04 said no rewrites, so that hop stays. A focused test now pins it (lets a never-themed profile inherit the last named-profile assign).

promoteUnanimousLegacy stays a write-on-read from resolve. It is idempotent. It no-ops when the legacy slot is set or when records disagree. A one-line why is on the helper. Leaving it on every resolve keeps Bot Mode hops and cold start on the same path.

[FalconOrtiz]

Awaiting maintainer. No merge from this side.

HEAD is still ada0347. GitHub reports MERGEABLE with mergeStateStatus BLOCKED. There are no review threads. watch-pr --status-only ended as BLOCKER status-query because the check rollup is empty.

I inspected the Actions tab. CI, Nix flake check, and Docker on this SHA are awaiting maintainer approval. Jobs have not started (0 jobs per run). That is not a Sonar failure and not a test failure. Approving those runs needs admin on NousResearch/hermes-agent. FalconOrtiz got HTTP 403.

This branch was rebased onto origin/main 8706517. Main's stored() / storedSkin() / committedName stayed. Named assign still mirrors persistString(legacy, value). promoteUnanimousLegacy is still write-on-read and now runs inside stored(). Local e2e smoke for this SHA is e2e_smoke_ok (C:\Users\falco\.cursor\orchestrate\hermes-agent\reports\e2e-smoke-101528-ada03473.md). There is no Sonar workflow in this repo, so excellency is N/A until checks exist.

Enough1122's LGTM nit still stands. Promotion persists on read. Pack agreement 2026-09-04 said nits are non-blocking and no rewrites. kvnloo's last-assign question was answered earlier. Last named assign is the intended hop. The four #101216 tests remain, including that hop.

origin/main is still 8706517 and is an ancestor of HEAD. GitHub reports no merge conflicts. I did not rebase again.

Approve the waiting workflows, then review and land if you want this in.

[FalconOrtiz]

Follow-up after rebase onto current main (ada0347390).

1d1411df9c split the old resolve helper into stored() / storedSkin(). Promotion now runs inside stored(), because ThemeProvider paints skin through storedSkin then skinPref.stored, not resolve. resolve is normalize(stored()), so Bot Mode hops still see the promoted global slot.

Last-assign hop is unchanged. Named assign still writes the global slot. The last-assign test still pins forge=dark then atlas=light, so a never-themed profile inherits light.

The earlier note that promotion lived only on resolve is stale after this rebase.