refactor(galgame): UI cleanup + capture chain (dxcam→mss→pyautogui) + dead-code cleanup (#1191)

* feat(galgame): add PyAutoGuiCaptureBackend, drop PrintWindow from default chain

Refactor screenshot backend chain to dxcam → mss → pyautogui as the new
fallback strategy. PrintWindow is intentionally removed from the default
chain because:

- It's a "render to DC" mechanism that often produces stale frames on
  DirectX/Unity games (already documented as `selected_legacy_fallback`
  in the previous detail label).
- Slower than BitBlt-based backends (mss/pyautogui).
- Inner implementation already falls back to BitBlt when PrintWindow
  itself fails, so its theoretical "capture occluded windows" advantage
  doesn't always materialize.

PrintWindow is still kept reachable in two narrower roles:

1. Explicit `selection="printwindow"`: chain becomes
   `[printwindow, dxcam, mss, pyautogui]` so power users who actually
   need occluded-window capture still have it as the first attempt with
   sane fallbacks if PrintWindow returns black/empty.

2. Smart-mode background target: still uses `[printwindow]` only since
   it's the only backend that can plausibly capture an occluded window
   (others read screen pixels). ocr_reader already emits
   `backend_not_suitable_for_background` warning when quality is poor,
   so the failure mode stays observable.

PyAutoGuiCaptureBackend is a thin wrapper over `pyautogui.screenshot()`
(already a main dep for brain/computer_use). On Windows it goes through
the same GDI path as mss; on macOS/Linux it shells out to platform
screenshot tools — slower than mss but battle-tested as a defense-in-depth
fallback.

New chain matrix:

| selection   | chain                                           |
|-------------|-------------------------------------------------|
| auto / dxcam| dxcam → mss → pyautogui                         |
| mss         | mss → dxcam → pyautogui                         |
| pyautogui   | pyautogui → dxcam → mss                         |
| printwindow | printwindow → dxcam → mss → pyautogui           |
| smart fg    | dxcam → mss → pyautogui                         |
| smart bg    | [printwindow] only (occluded-window capture)    |

Existing chain ordering tests updated.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* refactor(galgame-ui): remove rapidocr/dxcam install buttons, replace with bundled hint

PR #1188 removed the runtime install machinery for rapidocr/dxcam from
the Python side (HTTP routes, SDK actions, support helpers). The UI
side was left half-done: install buttons + JS handlers + i18n strings
were still in place. This commit completes the cleanup:

- static/main.js:
  - Drop RAPIDOCR_INSTALL_URL / DXCAM_INSTALL_URL constants
  - OCR_INSTALL_TABS = ['tesseract'] (was: rapidocr/dxcam/tesseract)
  - Drop rapidocr/dxcam entries from getInstallUIConfig() registry
  - Drop dependencySummaryItem rapidocr/dxcam branches (unreachable)
  - Drop installRapidOcr / installDxcam / restoreRapidOcr*Install*State
  - Drop case 'install_rapidocr' from action handler
  - Drop init promise.all entries for rapidocr/dxcam restore
  - Drop click event listeners for rapidocrInstallBtn / dxcamInstallBtn
  - activeInstallTab default 'rapidocr' → 'tesseract'
  - Add pyautoguiUseBtn click handler (new backend selection)

- static/main.js (renderRapidOcr / renderDxcam):
  Rewrite both functions to drop install-button DOM lookups and all
  install task state branches. Status banners (rapidocrPrompt,
  dxcamPrompt) now show:
  - Installed: ready/active state with detected_path + model_cache_dir
  - Not installed: bundled hint pointing dev to `uv sync --group galgame`
    or telling end-users to reinstall the packaged build
  - broken_runtime: "rebuild venv" hint for upgrade users with stale
    legacy installs
  Also added pyautoguiUseBtn config in renderDxcam alongside other
  capture-backend select buttons.

- static/index.html: drop <button id="rapidocrInstallBtn"> and
  <button id="dxcamInstallBtn">; add <button id="pyautoguiUseBtn">.

- i18n × 5 locales (zh-CN/en/ja/ko/ru): symmetric -25/+7 — drop
  install-action keys (action/retry/queued/success/failure/installing_*/
  install_kicker/failed_title/completed_refresh_title/install_first_title/
  missing_models_*) for both kinds; add bundled_hint, unavailable_title,
  pyautogui.use/.using/.title.

- integration test test_galgame_bridge_ui_routes.py:
  Drop assertions for now-removed JS strings (RAPIDOCR_INSTALL_URL,
  DXCAM_INSTALL_URL, restoreRapidOcrInstallState, restoreDxcamInstallState)
  and i18n key (ui.install.rapidocr.action). Re-target i18n bundle
  smoke check at ui.install.tesseract.action which still exists.

Plugin loads cleanly with `--group galgame` synced and without (rapidocr
status reports detail=missing in the latter case, which renders as the
bundled_hint banner instead of an install button). Chain ordering tests
still pass.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* chore(galgame): drop force_reload param + _purge_modules dead helper

Proposal A from PR #1188 cleanup discussion. The runtime install path
that used `force_reload=True` to evict cached `rapidocr_onnxruntime`
sys.modules entries is gone (HTTP routes deleted, install_rapidocr
function deleted). The remaining `inspect_rapidocr_installation`
callsite always passes `force_reload=False`, so the parameter and the
helper it gates are dead code.

- `load_rapidocr_runtime` drops `force_reload: bool = False` param
- `_purge_modules` function deleted
- Sole remaining callsite (inspect_rapidocr_installation) updated

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* chore(galgame): drop dead rapidocr_install_{manifest_url,timeout_seconds} cfg fields

Both fields fed only the runtime install machinery that PR #1188
deleted (HTTP fetch of install manifest JSON / subprocess timeout for
pip install). After install code removal, no consumer reads them.

Removed:
- `GalgameRapidOcrConfig.rapidocr_install_manifest_url`
- `GalgameRapidOcrConfig.rapidocr_install_timeout_seconds`
- Corresponding `_FIELD_MAP` entries in `GalgameConfig`
- Constructor kwargs in `service.py:_build_rapidocr_config`
- Test fixture entries in `test_ocr_pipeline.py`

Backward compat: old user cfg files with these keys → service.py loader
no longer calls `.get()` on them, so the dict entries are silently
ignored. No migration shim needed.

Kept (despite the misleading "install" prefix):
- `rapidocr_install_target_dir`: still actively used by ocr_reader as
  the runtime model cache root (rapidocr writes downloaded model files
  here). Renaming to `rapidocr_model_cache_root` belongs in a
  follow-up — too invasive (touches schema + UI + i18n).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(galgame): address codex P1+P2 on PR #1191

P1: load_rapidocr_runtime caller in ocr_reader.py was missed by the
force_reload removal commit. RapidOcrReader._ensure_runtime still
passed `force_reload=False`, which would raise TypeError on first OCR
runtime init (cache miss / warmup path). Drop the kwarg from the
caller to match the new signature.

P2: OCR_INSTALL_TABS shrinking to ['tesseract'] left two issues:
- HTML still rendered tab buttons for rapidocr/dxcam → clicking them
  coerced to tesseract via switchInstallTab fallback (UI inconsistency)
- revealCaptureBackendSettings() called switchInstallTab('dxcam') →
  same coercion to tesseract instead of focusing dxcam panel

Drop the rapidocr / dxcam install-tab buttons from index.html (their
banners are always visible now, no tab gating). Drop the dead
switchInstallTab('dxcam') call from revealCaptureBackendSettings —
expandAndScrollTo('dxcamPrompt') alone scrolls correctly without
needing to activate a (now nonexistent) tab.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(galgame-ui): re-route install_rapidocr action to bundled hint banner

Codex P2 on PR #1191: removing `case 'install_rapidocr'` from the
action handler switch left dangling action emitters. `withRapidOcrInstallAction()`
(general diagnosis flow) and the onboarding action button still produce
`install_rapidocr` action IDs. Source-install users without `--group
galgame` would see the action button, click it, and hit the default
"action unavailable" warning — broken UX.

Keep the action ID for backward compat with all emitter sites, but
change semantics + label:

- Re-add `case 'install_rapidocr'` to the action handler — now scrolls
  to `rapidocrPrompt` (where bundled_hint copy already explains
  reinstall packaged build / `uv sync --group galgame`).
- Action label: "一键安装 RapidOCR" → "查看 RapidOCR 启用提示" (and 5 locales)
- New flash key `ui.flash.rapidocr_hint_revealed` (5 locales) shown when
  the action triggers, telling user to follow the on-banner steps.

User who clicks now lands on the actionable hint instead of dead-end
warning. Action ID name "install_rapidocr" is misleading post-change
but fully internal; renaming would touch 5 client-side sites + i18n
key for cosmetic improvement only — not worth the diff.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(galgame-ui): renderPluginUnavailable safe for missing rapidocr/dxcam install config

Codex P1 on PR #1191: renderPluginUnavailable() iterates
['rapidocr', 'dxcam', 'tesseract', 'textractor'] and calls
getInstallConfig(kind) for each. Since rapidocr/dxcam were removed
from getInstallUIConfig() in commit e942d8e, this throws on the
first iteration — exactly the wrong moment, since this fallback
render runs precisely when the plugin is unavailable and diagnostics
are needed most.

Restructure the loop:
- Use a static PROMPT_LABELS table (rapidocr/dxcam labels are just
  brand strings, no need to round-trip through the install registry).
- Address banner DOM elements by `${kind}Prompt` directly (the prefix
  matches the kind for all 4).
- Guard `card`/`button` updates behind `kind in {'tesseract', 'textractor'}`
  — bundled rapidocr/dxcam no longer have install button/card DOM
  (removed in HTML cleanup).
- Defensive `if (banner)` early-continue: future-proof against
  any banner being removed without breaking the whole render.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(galgame-ui): address coderabbit review on PR #1191 (3 actionable)

3 valid concerns from coderabbit review of commits 99e1c1f + e942d8e:

1. **PyAutoGUI multi-monitor bug** (Major) — PyAutoGuiCaptureBackend
   used `pyautogui.screenshot(region=...)` which internally calls Pillow
   `ImageGrab.grab()` without `all_screens=True`. PyAutoGUI 0.9.54
   doesn't expose that flag, so secondary-monitor or negative-coordinate
   windows silently truncate to primary screen → OCR sees garbage.
   Switch to direct `ImageGrab.grab(bbox=..., all_screens=True)`.
   `is_available()` still gates on `import pyautogui` to keep the
   user-facing label honest. Comment explains why we bypass pyautogui's
   wrapper. all_screens=True is Windows-only in Pillow but harmlessly
   ignored on macOS/Linux.

2. **i18n PyAutoGUI casing** (Minor) — `ui.install.pyautogui.title`
   used lowercase `pyautogui` while `.use` and `.using` use brand
   `PyAutoGUI`. Was wrong in all 5 locales (en/ja/ko/ru/zh-CN, plus
   the main.js fallback string), not just the zh-CN line coderabbit
   flagged.

3. **Capture chain copy stale** (Minor) — `ui.install.capture_auto.title`
   and `ui.flash.capture_backend_settings_revealed` still listed
   "DXcam, MSS, PrintWindow" — missing PyAutoGUI which is now the
   third default-chain backend (PrintWindow moved to explicit-only +
   Smart background). Same issue across all 5 locales (coderabbit only
   flagged ru, but the rest had identical wording). Fold PyAutoGUI into
   the chain copy, note PrintWindow's narrower role.

i18n × 5 locale × 3 keys = 15 symmetric updates. ocr_reader.py change
is the substantive fix. Chain ordering tests still pass.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(galgame): address codex/coderabbit review on PR #1191 (3 actionable)

3 valid follow-ups from review of commit 87e7a65:

1. **Plugin-unavailable fallback leaves bundled banners hidden** (codex P2,
   main.js:2973): refreshAll() failure path rewrites RapidOCR/DXcam
   banner text but doesn't clear HTML `hidden`. Post-refactor
   switchInstallTab() only iterates ['tesseract'], so rapidocrPrompt /
   dxcamPrompt never get unhidden in the exact failure mode where the
   user needs the bundled-setup hint. Add explicit `banner.hidden = false`
   in the loop body.

2. **PyAutoGUI is_available crashes on non-ImportError** (codex P2 +
   coderabbit, ocr_reader.py:2505): pyautogui's import touches platform
   display state; in headless / WSL / missing-DISPLAY environments it
   may raise KeyError or RuntimeError instead of ImportError. With
   `except ImportError`, the exception bubbles up and aborts capture
   preflight. Widen to `except Exception` so backend probing degrades
   cleanly to "unavailable".

3. **Korean common values not localized** (coderabbit, ko.json:1010):
   `ui.common.none_value` was English `(none)`. While at it, also
   fix `ui.common.auto_value` which had the same issue. Now (없음) /
   (자동) — consistent with ja `(なし)/(自動)` and ru `(нет)/(авто)`.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(galgame): address coderabbit review on PR #1191 (2 actionable, 1 follow-up test)

1. **Major: Explicit PrintWindow chain semantics broken** (ocr_reader.py)
   `selection="printwindow"` chain was [printwindow, dxcam, mss, pyautogui]
   for ALL targets. PrintWindow's only reason to exist post-bundling is
   capturing occluded / background windows. If PrintWindow failed on a
   background target, silent fallback to dxcam/mss/pyautogui reads screen
   pixels — usually the occluding window, not the game — and OCR produces
   confident garbage from the wrong source.

   Fix: split _ordered_backends_for_target by target type, mirroring Smart
   mode's pattern:
   - Foreground target: keep [printwindow, dxcam, mss, pyautogui] (other
     backends also see the right window, fallback safe)
   - Background target: [printwindow] only (no silent screen-grab fallback)
   - Minimized: raise printwindow:target_window_minimized_for_capture

   New test test_win32_capture_backend_printwindow_strict_for_background_target
   locks in the foreground/background asymmetry.

2. **Minor: revealCaptureBackendSettings + install_rapidocr action don't
   expand collapsed parents** (main.js)
   `expandAndScrollTo('dxcamPrompt')` / `expandAndScrollTo('rapidocrPrompt')`
   only scroll. Both banners live inside `advancedSettings` + dependency
   `<details>` which default to collapsed. After scroll the user lands on
   a still-collapsed parent — the action effectively no-ops.

   Fix: prefix with navigateToInstallPanel(kind) which expands both
   containers (rapidocr/dxcam aren't in OCR_INSTALL_TABS so the tab-switch
   side of navigateToInstallPanel is a noop, but the expansion runs).

   Also synced JS fallback string for `capture_backend_settings_revealed`
   to match the i18n key (already includes PyAutoGUI from prior commit;
   the JS fallback was stale).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(galgame-ui): make navigateToInstallPanel scroll opt-out so banner scroll wins

Coderabbit minor on PR #1191: previous fix chained
`navigateToInstallPanel(kind)` + `expandAndScrollTo(banner)` to expand
collapsed parents before scrolling to the specific banner. But
navigateToInstallPanel queues `installSection.scrollIntoView()` in the
next animation frame — that fires AFTER expandAndScrollTo's
synchronous scroll, snapping the viewport back to installSection top
and hiding the banner the caller just landed on.

Add `scrollToSection: true` option to navigateToInstallPanel; the two
banner-targeted callers pass `false` to opt out of the queued scroll
while keeping the Advanced Settings + dependency `<details>`
expansion side effects.

- revealCaptureBackendSettings: navigateToInstallPanel('dxcam', { scrollToSection: false })
  + expandAndScrollTo('dxcamPrompt')
- case 'install_rapidocr': navigateToInstallPanel('rapidocr', { scrollToSection: false })
  + expandAndScrollTo('rapidocrPrompt')

Existing onboarding callers of navigateToInstallPanel keep the default
(scrollToSection=true) — they want the section-top scroll.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Hongzhi Wen <cartabio.coder1@gmail.com>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: 3 people

plugin/plugins/galgame_plugin/__init__.py

@@ -145,7 +145,7 @@ def _log_plugin_noncritical(logger: Any, level: str, message: str, *args: Any) -
 
 
 _OCR_BACKEND_SELECTIONS = {"auto", "rapidocr", "tesseract"}
-_OCR_CAPTURE_BACKEND_SELECTIONS = {"auto", "smart", "dxcam", "mss", "printwindow"}
+_OCR_CAPTURE_BACKEND_SELECTIONS = {"auto", "smart", "dxcam", "mss", "pyautogui", "printwindow"}
 
 
 def _migrate_legacy_capture_backend(value: object) -> object:
@@ -4055,7 +4055,7 @@ async def _poll_bridge_locked(self, *, force: bool) -> None:
                 and str(getattr(self._cfg, "ocr_reader_capture_backend", "") or "")
                 .strip()
                 .lower()
-                in {"smart", "dxcam", "mss", "printwindow"}
+                in {"smart", "dxcam", "mss", "pyautogui", "printwindow"}
             )
         )
         memory_reader_default_is_unavailable = (

plugin/plugins/galgame_plugin/i18n/ui/en.json

@@ -96,16 +96,6 @@
   "ui.install.memory_deps": "Memory Reader Dependencies",
   "ui.install.memory_desc": "Textractor is only used for the memory reader fallback path and does not affect OCR capture recognition.",
   "ui.install.running": "Installing in background...",
-  "ui.install.rapidocr.action": "Install RapidOCR",
-  "ui.install.rapidocr.retry": "Retry RapidOCR Installation",
-  "ui.install.rapidocr.queued": "Background installation task created. It will install the plugin-isolated RapidOCR runtime and stream progress via SSE.",
-  "ui.install.rapidocr.success": "RapidOCR installed",
-  "ui.install.rapidocr.failure": "RapidOCR installation failed",
-  "ui.install.dxcam.action": "Install DXcam",
-  "ui.install.dxcam.retry": "Retry DXcam Installation",
-  "ui.install.dxcam.queued": "Background installation task created. It will install the DXcam capture dependency and stream progress via SSE.",
-  "ui.install.dxcam.success": "DXcam installed",
-  "ui.install.dxcam.failure": "DXcam installation failed",
   "ui.install.tesseract.action": "Install Tesseract",
   "ui.install.tesseract.retry": "Retry Tesseract Installation",
   "ui.install.tesseract.queued": "Background installation task created. It will download Tesseract and language packs over HTTPS and stream progress via SSE.",
@@ -456,7 +446,6 @@
   "ui.field.memory_counts": "Memory Counts",
   "ui.install.rapidocr.using": "Using RapidOCR",
   "ui.install.rapidocr.force_title": "Force OCR Reader to use RapidOCR",
-  "ui.install.rapidocr.install_first_title": "Install RapidOCR first",
   "ui.install.ocr_auto.using": "OCR auto selection",
   "ui.install.ocr_auto.title": "Auto-select the OCR backend with RapidOCR first and Tesseract as fallback",
   "ui.status.unknown": "Unknown",
@@ -472,9 +461,6 @@
   "ui.install.rapidocr.ready_body": "RapidOCR has finished installing. When no SDK or valid memory text is available, it will take priority over Tesseract as the OCR Reader backend.",
   "ui.install.detected_path": "Detected path",
   "ui.install.model_dir": "Model directory",
-  "ui.install.rapidocr.kicker_missing_models": "Primary OCR backend missing models",
-  "ui.install.rapidocr.missing_models_title": "RapidOCR runtime exists, but the model state is incomplete",
-  "ui.install.rapidocr.missing_models_body": "The RapidOCR package exists, but the completed-install flag or model cache marker is missing. Reinstalling will warm-check and repair the state.",
   "ui.install.target_dir": "Target directory",
   "ui.install.rapidocr.kicker_broken": "Primary OCR backend error",
   "ui.install.rapidocr.broken_title": "RapidOCR runtime is broken or failed to import",
@@ -483,12 +469,7 @@
   "ui.install.rapidocr.not_ready_title": "RapidOCR is not ready yet",
   "ui.install.rapidocr.not_ready_body": "RapidOCR is now the internal primary backend for OCR Reader. After install, it will take priority over Tesseract, which remains only as a compatibility fallback.",
   "ui.install.expected_path": "Expected install path",
-  "ui.install.rapidocr.install_kicker": "RapidOCR installation",
-  "ui.install.rapidocr.installing_title": "RapidOCR is installing in the background",
-  "ui.install.rapidocr.installing_body": "The page receives live progress through SSE. When installation finishes, the plugin status will refresh and switch back to RapidOCR first.",
-  "ui.install.rapidocr.failed_title": "RapidOCR installation failed and can be retried",
   "ui.install.task_failed_retry": "The background install task failed. You can click the button again to retry.",
-  "ui.install.rapidocr.completed_refresh_title": "RapidOCR installation completed, refreshing OCR status",
   "ui.install.task_done_refreshing": "The install task has ended and the plugin status is waiting to refresh.",
   "ui.action.refresh_all": "Refresh All",
   "ui.action.debug_details": "View Debug Details",
@@ -499,7 +480,7 @@
   "ui.action.recalibrate_ocr": "Recalibrate OCR",
   "ui.action.line_details": "View Recognition Details",
   "ui.action.choice_advisor": "Switch to Auto Advance",
-  "ui.action.install_rapidocr": "Install RapidOCR",
+  "ui.action.install_rapidocr": "View RapidOCR setup hint",
   "ui.action.refresh_status": "Refresh Status",
   "ui.pending.locking": "Locking...",
   "ui.pending.clearing": "Clearing...",
@@ -533,7 +514,7 @@
   "ui.flash.sending_agent": "Sending to Agent...",
   "ui.agent.no_reply": "Agent returned no text",
   "ui.flash.agent_responded": "Agent responded",
-  "ui.install.capture_auto.title": "Auto-select the capture backend with DXcam first and MSS/PrintWindow as fallbacks",
+  "ui.install.capture_auto.title": "Default chain: DXcam first, MSS → PyAutoGUI as fallbacks (PrintWindow only as explicit selection or Smart background)",
   "ui.install.capture_auto.using": "Capture auto selection",
   "ui.install.compat_fallback": "compatibility fallback",
   "ui.install.current_plugin_python": "current plugin Python environment",
@@ -548,7 +529,6 @@
   "ui.install.printwindow.title": "Use Win32 PrintWindow fallback; DirectX/Unity games may return stale frames",
   "ui.install.dxcam.using": "Using DXcam",
   "ui.install.dxcam.force_title": "Force the capture backend to use DXcam",
-  "ui.install.dxcam.install_first_title": "Install DXcam first",
   "ui.install.dxcam.selected_waiting": "DXcam selected, waiting for the next OCR capture to confirm",
   "ui.install.dxcam.unsupported_title": "Automatic DXcam installation is not supported on this platform",
   "ui.install.dxcam.unsupported_body": "The DXcam capture backend is only for Windows desktop capture; this platform will continue using compatible capture options.",
@@ -561,11 +541,6 @@
   "ui.install.dxcam.kicker_not_ready": "Capture dependency not ready",
   "ui.install.dxcam.not_ready_title": "DXcam not detected",
   "ui.install.dxcam.not_ready_body": "After installing DXcam, the auto capture backend will prefer GPU Desktop Duplication for foreground game capture, reducing plugin-page occlusion and stale PrintWindow frames.",
-  "ui.install.dxcam.install_kicker": "DXcam installation",
-  "ui.install.dxcam.installing_title": "DXcam is installing in the background",
-  "ui.install.dxcam.installing_body": "The page receives live installation progress through SSE. After installation, the OCR capture backend status will refresh.",
-  "ui.install.dxcam.failed_title": "DXcam installation failed and can be retried",
-  "ui.install.dxcam.completed_refresh_title": "DXcam installation completed, refreshing capture status",
   "ui.install.tesseract.using": "Using Tesseract",
   "ui.install.tesseract.force_title": "Force OCR Reader to use Tesseract",
   "ui.install.tesseract.install_first_title": "Install Tesseract and required language packs first",
@@ -704,7 +679,7 @@
   "ui.flash.status_refresh_failed": "Status refresh failed. Try again later.",
   "ui.flash.debug_details_expanded": "OCR runtime debug details expanded.",
   "ui.flash.line_details_revealed": "Moved to current line recognition details.",
-  "ui.flash.capture_backend_settings_revealed": "Moved to capture backend settings. You can switch DXcam, MSS, or PrintWindow.",
+  "ui.flash.capture_backend_settings_revealed": "Moved to capture backend settings. You can switch DXcam, MSS, PyAutoGUI, or PrintWindow.",
   "ui.flash.focus_game_window": "Return to the game window. After it is foreground again, recognition will continue on the next refresh.",
   "ui.flash.action_unavailable": "This action is not available yet.",
   "ui.flash.refreshing_plugin_status": "Refreshing plugin status...",
@@ -1032,5 +1007,13 @@
   "ui.error.plugin_call_timeout": "Plugin call timed out",
   "ui.settings.saving_hint": "Saving...",
   "ui.common.auto_value": "(auto)",
-  "ui.common.none_value": "(none)"
+  "ui.common.none_value": "(none)",
+  "ui.install.rapidocr.bundled_hint": "RapidOCR is now bundled with the main program. If you're running the packaged build, please reinstall it; if you're running from source, run `uv sync --group galgame` and restart.",
+  "ui.install.rapidocr.unavailable_title": "RapidOCR unavailable — use the packaged build, or run `uv sync --group galgame` from source",
+  "ui.install.dxcam.bundled_hint": "DXcam is now bundled with the main program (Windows only). If you're running the packaged build, please reinstall it; if from source, run `uv sync --group galgame` and restart. The capture chain falls back to MSS / PyAutoGUI automatically.",
+  "ui.install.dxcam.unavailable_title": "DXcam unavailable — Windows only, requires packaged build or `uv sync --group galgame`",
+  "ui.install.pyautogui.use": "Use PyAutoGUI",
+  "ui.install.pyautogui.using": "Using PyAutoGUI",
+  "ui.install.pyautogui.title": "Use the cross-platform PyAutoGUI screenshot backend (slower on macOS/Linux)",
+  "ui.flash.rapidocr_hint_revealed": "Scrolled to the RapidOCR status banner. Follow the on-banner steps (reinstall packaged build / uv sync --group galgame)."
 }