speech-captionslisted

# Speech Captions (synced subtitles) Use this when the user wants **subtitles synced to speech** — not decorative text. Concern #1 is readability + sync, not animation. Captions are **structured text overlays grouped into a caption track** — one text overlay per cue, each with its own `startTime` / `duration`, all sharing a `caption.groupId`. There is no JS draw function to author or validate and no per-overlay code file. The fast path is `libi.generate_captions`, which builds the whole track in one call from the file's existing word timings. ## Workflow 1. **Ensure a transcript exists.** Captions need word-level timings. Prefer the `video-analysis` transcript for the spoken clip; if none exists, run the `audio-analysis` skill first (local Whisper → word-level timing). With only plain text available, sync will be approximate — state this to the user. 2. **Generate the caption track in ONE call:** ``` libi.generate_captions({ pieceId, fileId, // the video/audio file whose transcript drives the cues style: "cumulative", // a caption STYLE = reveal mode (see the style map below); default "cumulative" anchor: "bottom-center" // 3×3 numpad placement; default bottom-center }) ``` It reads the per-word timings (`analysis_get_audio_chunks` under the hood), builds readable, non-overlapping cues (lead + hold, width-budgeted, ≤2 lines), and creates one text overlay per cue sharing a per-file `caption.group