What problem does it solve? Screen recordings that scroll or pan through a document are hard to convert back into the original text: headings flash by, tables split across frames, and side comments are easy to miss. This Skill rebuilds the full document—outline, body, tables, code, images, and anchored comments—as an auditable Markdown or Obsidian note instead of a lossy summary. ## Core Features & Use Cases - Outline-First Recovery: Builds a complete title hierarchy with timestamps and a continuous, gap-free time partition before any body transcription, with a user confirmation checkpoint. - Evidence-Backed Transcription: Uses OCR only to locate text, verifies every line against original-resolution frames, and marks unreadable content as [画面不可辨] instead of guessing. - Visuals and Comments Handling: Extracts or rebuilds diagrams (Draw.io with PNG export), reconstructs illustrations with honest labeling, and stores side comments in a separate linked Obsidian note with stable block IDs. - Use Case: You recorded a scrolling session of a 40-page specification document with reviewer comments. The Skill produces a main Obsidian note with the full recovered text, a linked comments note, restored diagram assets, and a coverage report proving no time interval was skipped. ## Quick Start Use the video-document-recovery skill to reconstruct the full outline, body text, tables, and comments from my scrolling document recording and write the result as an Obsidian note.
