What problem does it solve? Turning video content into trustworthy written notes is error-prone: transcripts may be partial, automatic captions corrupt names and numbers, and it is easy to present creator claims as verified facts or invent timestamps. This Skill produces source-faithful, timestamped notes that make every quotation, claim, and coverage gap auditable. ## Core Features & Use Cases - Coverage-classified notes: Each video is classified as complete-transcript, partial-transcript, metadata-only, or unavailable, with omitted intervals and retrieval limits recorded. - Timestamp and claims ledgers: Timestamps are attached only with a known map and basis, and every claim preserves its speaker, locator, source quality, and what independent verification would need. - Compare mode: Multiple videos are synthesized against one common comparison frame, treating unequal transcript coverage as evidence asymmetry rather than a negative judgment. - Use Case: You have three conference talk recordings with auto-generated captions and want a single comparison of their positions on a topic, with exact quotations, timecodes, and a queue of claims ready for fact-checking. ## Quick Start Use sd-video-notes with videos set to the supplied URLs, mode=compare, and depth=standard to produce timestamped notes and a claims ledger for each video.
