refactor(main): move plugins, session actions and subtitle generation into assemblies (#327)

* refactor(main): build overlay geometry and pointer interaction in the overlay assembly

Inject Electron's screen (and the stats-window promotion) into the overlay geometry and
visible-overlay interaction runtimes so they load under bun test, then construct both in
createOverlayAssembly together with the overlay visibility composer and the mpv subtitle
suppression handlers. Replace the three regex tests that pinned the moved Linux input and
shape code with behaviour tests on the runtimes.

* refactor(main): move the autoplay-ready gate into the playback assembly

The autoplay-ready gate, its post-warm subtitle prime, the current-media tokenization gate and
managed local subtitle selection all read playback state, and playback signals through the
gate, so createPlaybackAssembly now builds them. The overlay reaches the gate through a getter.

Replace the last four regex guards with behaviour tests: the gate holds the release until the
visible overlay has measured the line and ignores untokenized lines during warmup (playback),
cached autoplay priming never paints raw first and the sidebar media path is tagged only after
prefetch loads (priming runtime), and dashboard mining overrides the Hachidori Anki server only
when the proxy is on (stats server runtime). main-wiring.test.ts is now empty and removed.

* refactor(main): move the plugin host, panel and windows into a plugins assembly

* refactor(main): move session action dispatch and renderer mpv commands into an assembly

createSessionActionsAssembly owns the session action dispatcher, the secondary-subtitle mode cycle, queueing the clipboard's video, and the mpv commands the overlay renderer sends. The runtime option cycle both paths used is now one function.

* refactor(main): move Whisper subtitle generation into an assembly

* refactor(main): resolve subtitle annotation options in a runtime module

The runtime-option-over-config resolution that the texthooker page and the MeCab startup decision share moves out of main.ts into createAnnotationOptions.

* test(main): cover a non-boolean runtime annotation option value

* docs: describe the plugins, session actions and subtitle generation assemblies

Update ownership for overlay geometry and pointer interaction, the autoplay-ready gate, and annotation options; point the overlay priming entry points at their new modules; the main-wiring regex debt is gone.

* build: export MpvClientEvents so declaration emit can name the plugins assembly's return type

docs/architecture/README.md

@@ -3,7 +3,7 @@
 # Architecture Map
 
 Status: active
-Last verified: 2026-05-23
+Last verified: 2026-11-01
 Owner: Kyle Yasuda
 Read when: runtime ownership, composition boundaries, or layering questions
 
@@ -30,14 +30,14 @@ Update checks and startup launcher migration share a serialized update-state sto
 
 - `src/main/` owns composition, runtime setup, IPC wiring, and app lifecycle adapters.
 - `src/main/boot/` owns boot-phase assembly seams so `src/main.ts` can stay focused on lifecycle coordination and startup-path selection.
-- `src/main/runtime/linux-overlay-mode-runtime.ts` owns Linux fullscreen mode state and window replacement. App cleanup cancels pending replacements; `main.ts` supplies window creation and subtitle refresh hooks.
+- `src/main/runtime/linux-overlay-mode-runtime.ts` owns Linux fullscreen mode state and window replacement. App cleanup cancels pending replacements; the overlay assembly supplies window creation and subtitle refresh hooks.
 - `src/core/services/` owns focused runtime services plus pure or side-effect-bounded logic.
 - `src/core/services/subtitle-generation*.ts` shares local whisper.cpp transcription, safe model downloads, and progress between the launcher and Electron. Optional dialogue mode retains both Silero-detected speech and other audible sections, omits confidently silent gaps, decodes passages independently, and restores original media timing. `src/main/runtime/subtitle-generation-runtime.ts` owns the overlay job lifecycle and only loads completed subtitles into the same media. For Jellyfin playback it reads the item's original-file stream (`toJellyfinOriginalStreamUrl`) into one reused output folder, where each run best-effort deletes earlier files and writes a uniquely named SRT; transcoded playback uses Jellyfin's audio index and only external references, since mpv's track indexes describe the transcode. Subprocess errors redact URL tokens; `src/shared/subtitle-generation*.ts` owns configuration, the multilingual model catalog, and IPC contracts. The overlay runtime retains a session model selection, validates picker requests through IPC, and keeps external model paths authoritative.
 - Subtitle model recommendations use bounded `nvidia-smi` and Whisper CUDA discovery probes in `subtitle-generation-acceleration.ts`. The overlay runtime caches results by executable path for 30 seconds and exposes acceleration status through the existing status IPC. Recommendations do not alter model selection or transcription arguments.
 - `subtitle-generation-reference.ts` ranks mpv's loaded text subtitle tracks, excludes signs/songs and forced references, and extracts timing hints with FFmpeg. The overlay and launcher snapshot references only for matching media, including active subtitle delays. Hints guide long-passage cuts with or without VAD; they never limit audio coverage or replace Whisper timestamps.
 - `src/main/runtime/subtitle-selection.ts` reads and validates mpv subtitle tracks and applies primary/secondary selections. Its opt-in session shortcut opens the shared overlay modal window, with renderer focus and subtitle suppression handled by the modal registry.
 - `src/shared/session-key-sequences.ts` rejects sequence prefixes reserved by single-key actions. The session-binding compiler reserves configured and built-in overlay keys; `src/main/runtime/session-bindings-runtime.ts` adds active mpv bindings and publishes the effective list to both the plugin artifact and the renderer through `session-bindings:changed`. mpv no-op `ignore` bindings do not reserve prefixes.
-- `src/core/services/plugins/` owns user plugins: discovery from `<config dir>/plugins/<id>/plugin.json` for ids enabled in `plugins`, the per-plugin `ctx`, guarded callbacks, and `exec` (no shell, AppImage `LD_LIBRARY_PATH` dropped). `main.ts` supplies the host's deps, attaches each new mpv client so plugin listeners survive client replacement, starts plugins from the app-ready path after logging is configured, and unloads them first in the `will-quit` cleanup, before any teardown. Plugins run in the main process with no sandbox. `src/plugin-api.ts` is the standalone public v1 type contract; internal mpv event and session-action types derive from it. Manifests may declare `apiVersion`, which is checked before loading code. Async activation is bounded and actions wait only for their own plugin. The host owns cached media snapshots and playback ids, including repeated loads of the same path. `ctx.signal` aborts as unloading starts; `ctx.onDispose` manages outside resources and late registrations are disposed immediately. Awaitable `ctx.mpv.request` uses the existing request/response transport. Plugin actions are the `runPluginAction` session action (keybinding command `__plugin-action:<id>/<name>`, CLI `--plugin-action <id>/<name>`); `runAction` waits for an in-progress load so a cold-start CLI action is not dropped. Each plugin gets `<config dir>/plugin-data/<id>/` as `ctx.dataDir`. `ctx.anki` carries the `card-mined` event with unique kept `noteIds` and per-original-note records. Sentence-card creation snapshots media and subtitle context before asynchronous work; the proxy snapshots before forwarding an upstream request. `AnkiIntegration` keeps provenance under original ids through merges and queued YouTube jobs; polling uses null provenance. The event is fed by `AnkiIntegration.setCardsReadyCallback`: polling, the AnkiConnect proxy, and sentence-card creation report a note once its enrichment finishes (or when they see it, if auto-update is off); `AnkiIntegration` holds a sentence card's report until field grouping finishes, and the pending YouTube media queue reports notes whose media it is still waiting on (directly or through the note field grouping merged them into), which is later than `recordCardsMined` (stats counts at add time). Stats-dashboard mining does not emit it. `ctx.anki.invoke` sends one action to the effective `ankiConnect.url` with no retries. `src/main/runtime/plugin-window-runtime.ts` owns plugin windows (`ctx.ui.openWindow`): one window per plugin + HTML path, raised above the overlay like the settings windows, destroyed on unload. `src/main/runtime/plugin-panel-runtime.ts` owns the single `plugin-panel` modal: a new panel aborts the previous handle's signal, updates broadcast to both overlay windows keyed by `panelId`, and the latest content is re-sent once the renderer acknowledges the open.
+- `src/core/services/plugins/` owns user plugins: discovery from `<config dir>/plugins/<id>/plugin.json` for ids enabled in `plugins`, the per-plugin `ctx`, guarded callbacks, and `exec` (no shell, AppImage `LD_LIBRARY_PATH` dropped). `src/main/assemblies/plugins.ts` builds the host. The playback assembly attaches each new mpv client so plugin listeners survive client replacement, the startup assembly starts plugins from the app-ready path after logging is configured, and the `will-quit` cleanup unloads them first, before any teardown. Plugins run in the main process with no sandbox. `src/plugin-api.ts` is the standalone public v1 type contract; internal mpv event and session-action types derive from it. Manifests may declare `apiVersion`, which is checked before loading code. Async activation is bounded and actions wait only for their own plugin. The host owns cached media snapshots and playback ids, including repeated loads of the same path. `ctx.signal` aborts as unloading starts; `ctx.onDispose` manages outside resources and late registrations are disposed immediately. Awaitable `ctx.mpv.request` uses the existing request/response transport. Plugin actions are the `runPluginAction` session action (keybinding command `__plugin-action:<id>/<name>`, CLI `--plugin-action <id>/<name>`); `runAction` waits for an in-progress load so a cold-start CLI action is not dropped. Each plugin gets `<config dir>/plugin-data/<id>/` as `ctx.dataDir`. `ctx.anki` carries the `card-mined` event with unique kept `noteIds` and per-original-note records. Sentence-card creation snapshots media and subtitle context before asynchronous work; the proxy snapshots before forwarding an upstream request. `AnkiIntegration` keeps provenance under original ids through merges and queued YouTube jobs; polling uses null provenance. The event is fed by `AnkiIntegration.setCardsReadyCallback`: polling, the AnkiConnect proxy, and sentence-card creation report a note once its enrichment finishes (or when they see it, if auto-update is off); `AnkiIntegration` holds a sentence card's report until field grouping finishes, and the pending YouTube media queue reports notes whose media it is still waiting on (directly or through the note field grouping merged them into), which is later than `recordCardsMined` (stats counts at add time). Stats-dashboard mining does not emit it. `ctx.anki.invoke` sends one action to the effective `ankiConnect.url` with no retries. `src/main/runtime/plugin-window-runtime.ts` owns plugin windows (`ctx.ui.openWindow`): one window per plugin + HTML path, raised above the overlay like the settings windows, destroyed on unload. `src/main/runtime/plugin-panel-runtime.ts` owns the single `plugin-panel` modal: a new panel aborts the previous handle's signal, updates broadcast to both overlay windows keyed by `panelId`, and the latest content is re-sent once the renderer acknowledges the open.
 - `src/renderer/` owns overlay rendering and input behavior.
 - `src/config/` owns config definitions, defaults, loading, and resolution.
 - `src/types/` owns shared cross-runtime contracts via domain entrypoints; `src/types.ts` stays a compatibility barrel.

docs/architecture/domains.md

@@ -3,7 +3,7 @@
 # Domain Ownership
 
 Status: active
-Last verified: 2026-10-31
+Last verified: 2026-11-01
 Owner: Kyle Yasuda
 Read when: you need to find the owner module for a behavior or test surface
 
@@ -23,19 +23,27 @@ Read when: you need to find the owner module for a behavior or test surface
 - App startup: app-ready sequence, headless known-word refresh, overlay startup prerequisites and
   overlay runtime bootstrap: `src/main/assemblies/startup.ts`
 - Overlay/window state: `src/core/services/overlay-*`, `src/main/overlay-*.ts`
-  Overlay windows, visibility, Linux fullscreen handling, and the overlay-window IPC handlers
-  (modal open/close, pointer interaction, content measurement): `src/main/assemblies/overlay.ts`
+  Overlay windows, visibility, geometry and pointer interaction, Linux fullscreen handling, mpv
+  subtitle suppression while the overlay shows, and the overlay-window IPC handlers (modal
+  open/close, pointer interaction, content measurement): `src/main/assemblies/overlay.ts`
   Overlay-hosted modal openers: `src/main/assemblies/overlay-modals.ts`
 - MPV runtime and protocol: `src/core/services/mpv*.ts`
-  mpv client creation, mpv event wiring, AniSkip, subtitle prefetch, and subtitle sidebar state: `src/main/assemblies/playback.ts`
+  mpv client creation, mpv event wiring, AniSkip, the autoplay-ready gate that releases mpv's
+  startup pause, managed local subtitle selection, subtitle prefetch, and subtitle sidebar state:
+  `src/main/assemblies/playback.ts`
 - Keyboard input outside the renderer (overlay, global and numeric shortcuts, session bindings):
   `src/main/assemblies/shortcuts.ts`
+- Session actions (the dispatcher behind shortcuts, controller buttons, CLI commands and plugin
+  actions), the secondary-subtitle mode cycle, and the overlay renderer's mpv commands:
+  `src/main/assemblies/session-actions.ts`
 - System tray: `src/main/assemblies/tray.ts`
 - Yomitan/Hachidori extension loading, parser runtime deps, and dictionary settings windows:
   `src/main/assemblies/yomitan.ts`
   Windows executable lookup and detached process creation are shared in `src/main/runtime/mpv-process.ts`. The Windows launcher and Jellyfin handlers retain their own playback and connection workflows.
 - Subtitle/token pipeline: `src/core/services/subtitle-*.ts`, `src/core/services/tokenizer*`, `src/core/services/tokenizer/`, `src/subsync/`
   JLPT and frequency annotation lookups: `src/main/jlpt-runtime.ts`, `src/main/frequency-dictionary-runtime.ts`, assembled by `src/main/assemblies/dictionaries.ts`
+  Annotation toggles (a cycled runtime option over its config value): `src/main/runtime/annotation-options.ts`
+  Whisper subtitle generation runtime, modal and IPC: `src/main/assemblies/subtitle-generation.ts`
   Subtitle text filters (`subtitleFilter`): `src/shared/subtitle-text-filter.ts`. Primary text is filtered at its two entry points, live `sub-text` in `mpv-protocol.ts` and parsed cues in `playback.ts`, so every consumer sees the same text and live/cue matching still works. A config change re-reads `sub-text` and re-parses the active track (the prefetch controller's parse key).
 - First-run setup: `src/main/runtime/first-run-setup-*.ts`, assembled by `src/main/assemblies/first-run-setup.ts`
 - Anki workflow: `src/anki-integration/`, `src/core/services/anki-jimaku*.ts`
@@ -55,7 +63,7 @@ Read when: you need to find the owner module for a behavior or test surface
   assembled by `src/main/assemblies/jellyfin.ts`
 - YouTube playback, subtitles, media cache, and in-app browser: `src/core/services/youtube/`,
   `src/main/runtime/youtube-*`, assembled by `src/main/assemblies/youtube.ts`
-- User plugins: `src/core/services/plugins/` (host, `ctx`, `exec`), `src/main/runtime/plugin-panel-runtime.ts` (panel modal), `src/main/runtime/plugin-window-runtime.ts` (plugin windows), `src/renderer/modals/plugin-panel*.ts`; examples in `assets/plugins/`
+- User plugins: `src/core/services/plugins/` (host, `ctx`, `exec`), `src/main/runtime/plugin-panel-runtime.ts` (panel modal), `src/main/runtime/plugin-window-runtime.ts` (plugin windows), assembled by `src/main/assemblies/plugins.ts`; `src/renderer/modals/plugin-panel*.ts`; examples in `assets/plugins/`
 - Window trackers: `src/window-trackers/`
 - Stats HTTP app: `src/core/services/stats-server.ts`, with route groups and shared route support
   in `src/core/services/stats-server/`

docs/architecture/subtitle-overlay-priming.md

@@ -3,7 +3,7 @@
 # Subtitle Overlay Priming
 
 Status: active
-Last verified: 2026-08-19
+Last verified: 2026-11-01
 Owner: Kyle Yasuda
 Read when: debugging subtitle state or blank Linux/X11 overlay windows when the visible overlay is shown or recreated
 
@@ -19,13 +19,14 @@ subtitles do not draw.
 
 ## Entry Points
 
-- `src/main.ts` calls `primeCurrentSubtitleForVisibleOverlay()` when manual visible-overlay show
-  paths run.
-- `src/main.ts` calls `restoreVisibleOverlayWindowShapeForShow()` before visible-overlay show
-  actions on Linux, and `resetVisibleOverlayInputState()` restores a full shape instead of applying
-  an empty shape.
-- `src/main.ts` also restores the Linux/X11 shape after applying mpv overlay bounds, so a newly
-  created 800x600 hidden Electron window cannot keep clipping after it is resized to mpv geometry.
+- The overlay assembly (`src/main/assemblies/overlay.ts`) calls
+  `primeCurrentSubtitleForVisibleOverlay()` when manual visible-overlay show paths run.
+- It calls `restoreVisibleOverlayWindowShapeForShow()` before visible-overlay show actions on
+  Linux, and `resetVisibleOverlayInputState()` restores a full shape instead of applying an empty
+  shape (both in `src/main/runtime/visible-overlay-interaction-runtime.ts`).
+- `src/main/runtime/overlay-geometry-runtime.ts` also restores the Linux/X11 shape after applying
+  mpv overlay bounds, so a newly created 800x600 hidden Electron window cannot keep clipping after
+  it is resized to mpv geometry.
 - `primeCurrentSubtitleForVisibleOverlay()` delegates to
   `primeVisibleOverlaySubtitleFromMpv()` in `src/main/runtime/current-subtitle-snapshot.ts`.
 - `restoreVisibleOverlayWindowShapeForShow()` delegates to `restoreLinuxOverlayWindowShape()` in
@@ -54,7 +55,7 @@ Both `onSubtitleChange` and `refreshCurrentSubtitle` pause `subtitlePrefetchServ
 the matching `subtitleProcessingController` method, giving the visible overlay priority over
 background prefetch work. Prefetch is not re-centered here: restarting the run per line
 (`onSeek`) discarded the in-flight tokenization every time the subtitle changed, so only real
-seeks restart it (see `onTimePosUpdate` in `src/main.ts`).
+seeks restart it (see the time-pos handling in `src/main/assemblies/playback.ts`).
 
 On an uncached autoplay prime the raw payload is emitted here and reported to the controller with
 `notePlainSubtitleEmitted`, so the controller skips its own plain emit for that line and the