@joewinke/jatui 0.34.1 → 0.35.0

package/CLAUDE.md

@@ -122,6 +122,7 @@ Item-by-item review surfaces (MOS review-ux, design mos-z8vgx). One `VerdictOpti
 | `VideoPlayer` | **The default video surface — reach for this over a bare `<video controls>`.** One timeline band carrying a waveform-envelope scrubber, chapter flags (top edge), bookmark diamonds (bottom edge), diarized speaker colouring and scroll-to-zoom, plus an optional transcript pane (click to seek, active line auto-scroll, solo/skip per speaker, `T` toggle) and a "Talk time" legend (2+ speakers only; one voice gets no legend and no per-row name). **`kind="audio"` renders no stage**, and **`transport="band"`** (the audio default) drops the native `<video controls>` bar for a transport row on the band (play/pause, `now / total`, speed, mute + volume, fullscreen for video), so the band is the ONE timeline. **Every intelligence prop is optional** — a bare mp4 still gets the scrubber, hover-time card, wheel volume and the full keyboard map (Space/K, arrows/J/L, vol/mute, −/= speed + Backspace reset, </> frame step, [/] chapters, B bookmark, ? help overlay), and each pane lights up as its data arrives (nothing shifts layout when diarization lands late). Wheel means volume over the video (shift = seek, alt = speed) and zoom over the band. Internals live in `src/lib/video/` (band/keys/rate/deepLink/settings + contracts for audio/night/presence — see that README; gap audit vs the extension: `VIDEO-PLAYER-GAP-AUDIT.md`). | `src`, `kind?` (`'video' \| 'audio'`), `transport?` (`'native' \| 'band'`), `chapters?`, `speakers?`, `transcript?`, `bookmarks?`, `waveform?`, `talkTime?`, `startAt?`, `poster?`, `captionsSrc?`, `crossorigin?` (`'anonymous'` — what frame-reading features need), `syncUrlTime?` (`?t=` deep-link, opt-in), `transcriptPosition?` (`'below' \| 'side'`), `onTimeUpdate?`, `onAddBookmark?`, `onBookmarkContext?` |
 | `MediaCapture` | One-tap audio/selfie-video/screen(+mic) recording — a host-agnostic PRIMITIVE. Owns permission requests, MIME negotiation, the record/stop state machine, live self-preview, and mixes screen+mic audio into one track via AudioContext. Never uploads anything — hands the finished clip to `onCapture` as a Blob and the host decides what to do with it. `onCapture` may be async; the component shows "processing" until it resolves. | `onCapture: (result: MediaCaptureResult) => void \| Promise<void>`, `modes?` (default all three), `maxDurationMs?` (default 120000), `label?`, `hint?`, `compact?`, `disabled?` |
 | `MultimodalCaptureField` | "Upload a file, or record yourself going over it" as one input surface — composes `FileDropzone` (capped to one file) with `MediaCapture`. File and recording are independent (either, both, or neither); neither half uploads anything, same "validate and hand back" contract as `FileDropzone`. | `onFileSelected?`, `onFileRemoved?`, `onCapture?`, `onRecordingRemoved?`, `accept?`, `maxFileSize?`, `modes?`, `maxDurationMs?`, `file?`/`fileUrl?`/`fileName?`, `recordingSummary?` |
+| `ListenControl` | "Listen to this page" — a compact pill that expands into a playback bar over the host's PER-SENTENCE audio (TTS): play/pause, −15s/+15s and prev/next sentence, speed (0.75–2×), voice picker, a document-wide progress bar with chapter ticks (a hatched tail marks the not-yet-synthesized part), time, close. The host supplies an async `source` (sentences/paragraphs with ids → audio URLs, streamed page by page so playback starts on the first chunk while the rest synthesizes; pages merge by id so late URLs fill in) and gets `onSegmentChange(id)` at every boundary for highlight + scroll-follow. Media Session (lock-screen/headset controls, one document-wide position), keyboard scoped to the player (Space/K, ←/→ sentences, Shift+←/→ ±15s, [ / ] speed, Escape), speed+voice persisted in localStorage, `bind:api` for hosts wiring their own global keys, reduced-motion aware, token-only theming, fixed sticky mini-bar on narrow screens. Internals in `src/lib/listen/` (one timeline across per-sentence files — `buildTimeline` offsets; see `video/` for the doctrine). Demo synthesizes real WAVs locally. | `source`, `voices?`, `voiceId?` (bindable), `chapters?`, `title?`, `triggerLabel?`, `open?` (bindable), `preload?`, `estimateSec?`, `storageKey?`, `onSegmentChange?`, `onOpenChange?`, `api?` (bindable), `class?`, `children?` (custom trigger) |
 
 **`VideoPlayer` vs `TranscriptPlayer`:** VideoPlayer is the generalization — it does everything TranscriptPlayer does plus the timeline band, and its props are structurally compatible with TranscriptPlayer's (`TranscriptSegment[]`, `TranscriptChapter[]`, `TranscriptBookmark[]`, `TranscriptSpeakerTalkTime[]` all pass straight through). `TranscriptPlayer` stays exported and unchanged for the existing meeting-recordings viewer; use `VideoPlayer` for anything new.