feat: 优化录音详情工作区与应用图标

CHANGELOG.md

@@ -7,6 +7,26 @@ and this project follows [Semantic Versioning](https://semver.org/spec/v2.0.0.ht
 
 ## [Unreleased]
 
+### Changed
+
+- Reorganized recording details into a full-height reading workspace with a
+  compact contextual toolbar, document/version selectors, bottom audio player,
+  a collapsible recording list controlled from the left of the detail title,
+  and an on-demand generation-details drawer. Existing
+  generation, regeneration, AI revision and transcription dialogs retain their
+  fields and behavior.
+
+### Fixed
+
+- Unified the sidebar and tray with Nota's packaged application icon. Idle
+  shows the app mark; recording adds a lower-right red badge, while pause and
+  capture warnings retain distinct status badges and existing tray controls.
+- Unified AI generation close and JSON disclosure icons with Lucide. Generation
+  dialogs and the details drawer now support backdrop dismissal without closing
+  on text-selection drags; generation submission consistently blocks dismissal.
+- Kept AI generation form scrolling inside the rounded dialog shell, with the
+  header and actions visible and no scrollbar covering the corners.
+
 ## [0.9.0] - 2026-09-02
 
 ### Changed

docs/ai-documents.md

@@ -110,8 +110,8 @@ contain credentials, prompt bodies, transcript content, or model output.
   Provider-controlled JSON verbatim but Nota never adds runtime credentials to
   it. Both snapshots are loaded by React only when the user opens **Generation
   details**.
-- The document preview has top-level **Document** and **Generation details**
-  tabs. Generation details shows normalized usage plus separate request and
+- The document preview keeps the body visible and opens **Generation details**
+  in a right-hand drawer. Generation details shows normalized usage plus separate request and
   response JSON trees. Historical versions created before snapshot persistence
   explicitly report that their original JSON is unavailable.
 

docs/architecture.md

@@ -76,6 +76,39 @@ flowchart LR
 | Configured ASR service | Model inference and server-side processing | Local recording ownership |
 | Configured LLM provider | Explicitly requested text generation | Local Markdown, SQLite, or recording ownership |
 
+### Recording Detail Presentation
+
+`RecordingsWorkspace` owns selected-record presentation, native audio playback,
+transcription actions, focus mode, and the existing speaker/transcription
+dialogs. `AiDocumentsPanel` remains the sole owner of AI workspace reads,
+generation drafts, request estimates, version events, file actions, and all
+three generation-dialog modes. Its toolbar, Markdown reader, and generation
+details drawer are presentation components, not additional IPC owners.
+
+The AI panel mounts on first visit and remains mounted while its tab is hidden.
+The audio element, record list, and transcript scroller also retain their DOM
+identity across tab/focus changes. Markdown reading positions are keyed by
+version in transient component state; they are not stored in SQLite or settings.
+Leaving the workspace clears this presentation state. Recording changes retain
+the existing AI reset behavior and asynchronous identity guards.
+
+```mermaid
+flowchart LR
+    A[App: workspace shell] --> B[RecordingsWorkspace: audio and detail tabs]
+    B --> C[AiDocumentsPanel: selected document and version]
+    C --> D[Toolbar: user intent callbacks]
+    C --> E[Reader: Markdown and transient scroll positions]
+    D --> F[Open generation details drawer]
+    F --> G[Read persisted details for selected version]
+    G --> H[Apply only if request and version are current]
+    D --> I[Existing create / regenerate / revise dialog]
+    I --> J[Existing preview, validation and explicit generation IPC]
+```
+
+Opening tabs, reading documents, and inspecting persisted details do not invoke
+LLM generation. No new backend commands, schemas, credentials, or network paths
+are introduced by the workspace layout.
+
 ### Settings Ownership and Persistence
 
 `App.tsx` owns the canonical settings and Provider summaries shared with the
@@ -253,6 +286,26 @@ right corner. Content beyond the upper bound remains scrollable, so DPI,
 accessibility text scaling, long target titles, and error details cannot hide
 the decision buttons.
 
+## Application Identity and Tray Rendering
+
+The frontend sidebar imports the packaged app-icon PNG as a build asset, and
+Rust embeds the 32 px variant with `tauri::include_image!`. Neither path needs
+runtime filesystem access, IPC image payloads, or an extra image dependency.
+The tray compositor always starts from the original app mark, adding only a
+lower-right anti-aliased badge when the state needs one. Returning to idle
+restores the original pixels; badges never accumulate across transitions.
+Recording preparation/finalization, pause and capture-warning distinctions
+remain visible. Existing state deduplication, main-thread tray updates, menus,
+tooltips and recording control ownership are unchanged.
+
+```mermaid
+flowchart LR
+    S["Recording state + pending capture decision"] --> B["Choose no badge / recording / paused / warning"]
+    I["Packaged 32 px application icon"] --> C["Compose corner badge from original pixels"]
+    B --> C
+    C --> T["Update tray icon on main thread"]
+```
+
 ## Live Microphone Switching
 
 An active recording may replace or disable only its microphone source. The

docs/design-system.md

@@ -126,6 +126,17 @@ Lucide remains Nota's product icon set. Use 14 px icons in compact controls,
 buttons need an accessible name and an `AppTooltip`; visible text buttons do
 not need a duplicate tooltip.
 
+Application identity is not a Lucide action icon. The sidebar imports the
+packaged `src-tauri/icons/128x128.png` directly, with no replacement glyph,
+extra background or clipped silhouette. Installer, window and tray icons use
+the same packaged microphone artwork; the tray embeds `icons/32x32.png`.
+Idle/completed/recovering tray states show the unmodified app mark. Recording,
+preparing and finalizing add a lower-right recording-red badge; paused adds
+a paused-amber badge. Interrupted/error states and pending capture decisions
+show a warning-amber badge, with pending decisions taking priority. Badges have
+an anti-aliased light separator and use native equivalents of the existing
+status color tokens. Existing tooltip/menu text supplies the precise state.
+
 | Control | Standard |
 |---|---|
 | Compact icon button | 28 px minimum visual target, 14 px icon |
@@ -186,6 +197,39 @@ are removed so the settings hierarchy owns the available height.
 Settings uses existing semantic surface, border, status, spacing, and focus
 roles. The workspace geometry does not introduce a new visual token role.
 
+### Recording Detail Workspace
+
+The recording library uses the full application height with the main sidebar
+retained and the redundant global header/footer removed. The record list is a
+280 px track (240 px at widths up to 1100 px). The icon button before the detail
+title toggles only that track, with matching tooltip and accessible labels
+“折叠录音列表” / “展开录音列表”. It exposes the list's expanded state and remains
+available when the list is hidden.
+The compact title, detail tabs, contextual toolbar, reading region, and native
+audio player follow DOM order. The list and reading region scroll independently;
+the outer detail pane must not scroll. Playback remains mounted across tabs.
+
+AI documents use shared select styling for document/version selection instead
+of a permanent nested document sidebar. AI revision and copying stay directly
+available; low-frequency file and generation actions use a keyboard-accessible
+disclosure. Generation details open in a right-hand native dialog drawer with
+focus containment and restoration. Generation configuration dialogs retain
+their original fields and controls. Their rounded outer shell clips overflow;
+only the active form/request panel scrolls, leaving the title, tabs and footer
+visible. Scrollbars stay inside the shell padding, clear of its rounded corners.
+Close controls use Lucide `X`; JSON tree disclosure slots use the bundled Lucide
+chevron SVG as a semantic-color CSS mask without replacing tree keyboard behavior.
+Both generation overlays dismiss on primary clicks that start and end on their
+backdrop. Inside clicks and text-selection drags must not dismiss them; generation
+submission disables close, cancel and backdrop dismissal together.
+
+At 1280×800 and 980×640, normal completed-document content targets at least
+70% and 60% of workspace height respectively. Necessary warnings may reduce
+that area. Controls must not wrap into multiple toolbar rows, and long titles,
+models, and paths must not cause page-level horizontal overflow. Use existing
+semantic tokens; do not shrink body type to achieve density. The playback
+footer retains the local-recording/network-use reminder.
+
 ## Change and Review Checklist
 
 For every visual change:

docs/testing.md

@@ -122,8 +122,8 @@ Automated client coverage must include:
 
 - recording-list context menus exposing **Open containing folder** only for
   list items while preserving the three-action detail overflow menu;
-- one sticky recording-detail control region containing the audio player and
-  transcription actions, with the transcript body outside that region;
+- compact recording-detail controls above an independently scrolling transcript,
+  with a stable native audio player below the body;
 - transcript segment timestamps rendered as zero-padded `HH:MM:SS`, including
   meetings longer than one hour;
 - valid batch protocol v1 capability parsing;
@@ -267,7 +267,7 @@ Automated coverage for AI documents must include:
 - credential-free request JSON matching the submitted body, successful raw
   response JSON and normalized usage persistence, nullable-column migration
   for historical databases, and on-demand IPC reads;
-- keyboard-operable Document/Generation details and Request/Response tabs,
+- a keyboard-operable Generation details drawer and Request/Response tabs,
   collapsible syntax-highlighted JSON, copy actions, long-value wrapping, and
   explicit unavailable states for historical versions;
 - recording deletion preserving Markdown by default and deleting exact linked
@@ -374,6 +374,48 @@ Real-model and hardware acceptance results should record software versions,
 model id, device type, audio duration, and pass/fail observations. They must not
 include the recording or transcript in ordinary logs or committed artifacts.
 
+## Recording Detail Layout Acceptance
+
+### Application Icon Acceptance
+
+- Confirm the sidebar microphone artwork matches the installed Start menu and
+  window icon. It must not render a replacement Lucide action glyph.
+- Check the Windows tray on light/dark taskbars at 100%, 125%, 150% and 200%
+  scaling: idle uses the app mark only; preparing/recording/finalizing add a red
+  lower-right badge. Pause uses amber; capture reminders/errors use warning
+  amber. Check tooltip/menu distinctions and pending-decision priority.
+- Finish recording and confirm the unmodified idle icon returns. Tray show,
+  pause/resume, stop/save and capture-prompt re-entry must still work.
+- Pixel tests verify the badge leaves the rest of the icon unchanged and that
+  idle states exactly match the embedded asset. Browser/pixel tests do not
+  replace this Windows system-tray check.
+
+### Layout and Dialogs
+
+- Verify the full-height recording workspace at 1280×800, 980×640 and Windows
+  125% scaling: persistent app navigation, independent list/body scrolling,
+  bottom playback controls, and no horizontal page overflow.
+- Check focus mode, keyboard tab switching, overflow actions, drawer Escape and
+  focus restoration. Tab/focus changes must not replace audio or reset playback
+  and the current record's reading positions. Recording-active warnings and the
+  return-to-recorder action remain available.
+- Compare create, regenerate and revise dialogs with the pre-layout baseline:
+  all fields, settings/request tabs, Provider/model linkage, token estimates,
+  budget errors and submit/cancel states must remain unchanged. Assert complete
+  generation payloads, including the revise source-version ID.
+- Regress generation progress/cancel, missing/externally modified Markdown,
+  file relinking, stale responses, and on-demand request/response inspection.
+  Retain all existing playback, speaker, export and transcription tests.
+- Verify generation dialogs and the details drawer close on backdrop clicks,
+  but not inside clicks, cancelled pointers, secondary buttons, or text-selection
+  drags ending outside. Submission blocks close/cancel/backdrop dismissal. Check
+  drawer focus restoration and JSON keyboard expansion after the Lucide icon
+  adaptation. At both viewport sizes, scroll the complete form and confirm
+  header/footer remain visible and scrollbars never cover the shell's corners.
+- Visual fixtures must be synthetic and must not call real model services.
+  Browser viewport checks do not replace real Windows/WebView2 scaling and
+  audio-device acceptance.
+
 ## Documentation Verification
 
 Documentation-only changes do not require rebuilding the application unless