BooksWriterStudio remake

name: BooksWriterStudio remake overview: Remake books-writer as a new Avalonia app BooksWriterStudio under novolis-apps/src/, consuming only published Novolis packages plus app glue—no Python and no dotnet run --file scripts. Extend Novolis.Markup.Manuscript (catalog/diagnostics/book PDF) and Novolis.Audio.Voice.Manuscript (TTS pipeline, concat MP3 / M4B assemble, selected-text speech preview); Manuscript Studio stays as the multi-mode editor. todos:

• id: markup-catalog-doctor-pdf

content: Extend Novolis.Markup.Manuscript with Catalog, Doctor, Book/Reference PDF + tests; publish to GPR status: completed

• id: voice-pipeline-assemble

content: Extend Novolis.Audio.Voice.Manuscript with synthesizer, VoiceMap, pipeline, ConcatMp3 + M4B assemble, verifier, and selected-text speech preview; publish to GPR status: completed

• id: app-scaffold

content: Create BooksWriterStudio Avalonia app, Directory.Packages.props + slnx wiring, workspace/editor/session glue status: completed

• id: app-publish-scm

content: Wire publish jobs (PDF/audio), selected-text audio preview, JobQueuePanel, diagnostics, git checkpoint, recovery/conflict/search dialogs status: completed

• id: verify-nuget-build

content: verify-nuget-only.ps1 + nuget.org/github restore/build of BooksWriterStudio status: completed isProject: false

Decisions locked

• App: new `novolis-apps/src/BooksWriterStudio/` (parity remake). Leave `ManuscriptStudio` alone.
• Audio: extend `Novolis.Audio.Voice.Manuscript` (not a separate Audiobook package). Assemble outputs: concatenated MP3 and/or chapterized M4B.
• No scripts: zero dotnet run --file, zero Python/edge-tts CLI, zero ffmpeg process. TTS via EdgeTtsClient library; Git may still spawn git through `Novolis.IO.Git` (system tool, not a script).

Architecture

flowchart TB
  app[BooksWriterStudio glue]
  studio[Avalonia.Studio Controls Markdown]
  ms[Markup.Manuscript Catalog Diagnostics PDF]
  voice[Voice.Manuscript plan pipeline assemble]
  edge[Voice.EdgeTts]
  io[IO Paths Recovery Watching Git]
  app --> studio
  app --> ms
  app --> voice
  app --> io
  voice --> edge

App owns UX orchestration only. Domain/publish logic lives in packages published to GitHub Packages, then referenced as 2026.1.*.

Reuse Avalonia lifts already in tree: ChoiceDialog, FilteredPickerDialog, MarkedListBox, JobQueuePanel / IJobQueueRow, StudioFocusMode (`generic_avalonia_lifts` plan).

Package work (publish before app wires)

1. Extend `Novolis.Markup.Manuscript` — Catalog

Lift from Manuscript Studio’s internal `ContentCatalog` into the package (do not add a second Catalog package).

Public API

Behavior: same layout as books repo / Studio (content/series/{id}/books/{id}/chapters/*.md, appendices, references, standalone content/books/{id}).

Tests: fixture tree with 1 series × 2 chapters + 1 standalone book; assert ids, titles, order.

2. Extend `Novolis.Markup.Manuscript` — Diagnostics

Replace book-tool doctor --json.

Public API

Checks (v1): missing book.yaml/series.yaml; chapter file missing/empty; duplicate chapter stems; heading title mismatch; unreadable markdown; orphan refs under references/; chapter_order_from_heading vs filename order conflicts.

Tests: fixture with planted errors; assert codes.

3. Extend `Novolis.Markup.Manuscript` — Book PDF

Replace compile-book.cs / compile-reference.cs process shell-outs. Prefer composing `MarkdownPdfExporter` (add PackageReference from Manuscript → Markdown.Rendering).

Public API

Tests: export tiny fixture book to bytes; assert non-empty PDF header %PDF.

4. Extend `Novolis.Audio.Voice.Manuscript` — pipeline + assemble

Keep existing SpeechPlanner / SpeechPlan / ManuscriptSpeechOptions. Add synthesis orchestration and assemble in this package.

New dependencies (nuget.org, allowed): Novolis.Audio.Voice.EdgeTts (GPR) for TTS; for M4B: NAudio MediaFoundation AAC encode (Windows) + small in-package ISOBMFF chapter writer (no ffmpeg).

Public API

Speech preview behavior (library)

• Input: raw editor selection (or current line/word if selection empty — app chooses; library accepts a non-empty string).
• Apply SpeechPlanner.ApplyPronunciation + light strip of markdown markers from the snippet (not full chapter planning).
• Use the same ManuscriptVoiceSettings as audiobook generation (voice/rate/pitch/volume).
• PreviewAsync(text, settings, ct) synthesizes then plays; a second call cancels the previous playback/synthesis.
• Cap preview length (e.g. 4k chars) with a clear exception so accidental whole-chapter selects do not hammer Edge TTS.

Pipeline flow

sequenceDiagram
  participant App
  participant Pipe as ManuscriptAudiobookPipeline
  participant Plan as SpeechPlanner
  participant TTS as IManuscriptSynthesizer
  participant Asm as AudiobookAssembler
  App->>Pipe: GenerateAsync(book, options)
  loop chapters
    Pipe->>Plan: Create(markdown)
    Pipe->>TTS: segments to chapter MP3
  end
  Pipe->>Asm: ConcatenateMp3 and/or WriteM4b
  Pipe-->>App: AudiobookResult + manifest

Tests: planner hash stability (existing); concat of two tiny MP3 fixtures → length increases; verifier fails on missing chapter; preview cancels prior run / rejects over-limit text (fake synthesizer + null/spy player); M4B test gated [Fact(Skip=...)] or Windows-only if MediaFoundation required.

Package description update: planning + chapter TTS orchestration + MP3/M4B assemble + selection speech preview.

App: `BooksWriterStudio`

Path: `novolis-apps/src/BooksWriterStudio/`

Shape: Avalonia 12 WinExe, Microsoft.Extensions.Hosting DI (same pattern as Manuscript/Concept). MVVM with CommunityToolkit.Mvvm. No process runners for catalog/doctor/PDF/audio.

PackageReferences

• Novolis.Avalonia.Studio, .Controls, .Markdown
• Novolis.Markup.Manuscript (catalog/doctor/PDF/metadata/word count)
• Novolis.Audio.Voice.Manuscript, Novolis.Audio.Voice.EdgeTts (preview + publish TTS)
• Novolis.IO.Paths, .Recovery, .Watching, .Git
• nuget.org: Avalonia stack, AvaloniaEdit, YamlDotNet, WeCantSpell.Hunspell (spell — app glue)

Add floating versions to `Directory.Packages.props`; register project in `Novolis.Apps.slnx`.

Feature map (glue only)

Layout (parity)

Three-column writing chrome inspired by books-writer (not Manuscript’s preview-first rail): chapter list | AvaloniaEdit | context tabs (Metadata, Search, Diagnostics, Publish, SCM). Editor chrome includes Speak selection / Stop next to save/zoom. Dialogs: Settings, Goto, Reference picker, Conflict, Recovery via Controls package.

Out of scope for v1

• Pass start/finish / revision tags UI (APIs exist on GitRepositoryService; wire later)
• Mermaid/StarMap views (stay in Manuscript Studio)
• Generic-markdown / Concept modes
• Installer/release workflow wiring (follow-up; can add later per `release.md`)

Delivery order

1. Markup.Manuscript Catalog + Diagnostics + Book PDF → unit tests → pack/publish GPR
2. Voice.Manuscript synthesizer + pipeline + ConcatMp3 + M4b + verifier + VoiceMap + ManuscriptSpeechPreview → tests → publish GPR
3. BooksWriterStudio app scaffold + editor/session/IO wiring
4. Publish panel + audio/PDF jobs + selection audio preview
5. Polish: spell, focus, SCM checkpoint, settings persistence
6. pwsh -File novolis-governance/scripts/verify-nuget-only.ps1; dotnet restore/build apps with nuget.org + github only

Done criteria

• Open a books-layout workspace, edit/autosave a chapter, recover after simulated crash, see doctor findings, checkpoint git, export book PDF, generate chapter MP3 and assemble concat MP3 and M4B, and preview selected text with the configured voice — all in-process libraries, no Python/ffmpeg/dotnet run --file.