What problem does it solve? Turning a teaching video, tutorial, or screen recording into a reusable Agent Skill normally requires manually transcribing steps, extracting frames and clips, and writing instructions from scratch. This Skill automates that conversion by perceiving the recording as a time-aligned audio-visual timeline and producing a complete SKILL.md with multimodal assets and test cases. ## Core Features & Use Cases - Video Perception: Reads a teaching video as one audio-visual timeline, building a timestamped event log (video_events.md) that captures steps, tools used, and presenter caveats. - Multimodal Asset Management: Plans, extracts, annotates, and registers frames, clips, and audio segments with an asset_manifest.json so consumers can verify their own results against the demonstration. - Three-Layer Verification: Runs structural validation (L1), asset pre-checks like dedup and orphan detection (L2), and single-arm task evaluation with executor and grader subagents (L3), aggregated into benchmark reports. - Description Optimization: Runs an optimization loop that rewrites the skill's description field against labeled trigger queries to improve triggering accuracy. - Use Case: A user records themselves configuring a dashboard in a BI tool and says "turn this tutorial into a skill." The Skill perceives the recording, drafts the SKILL.md, extracts key frames, runs test prompts, and delivers a validated, reusable skill. ## Quick Start Use the omni skill creator to turn this screen recording into a reusable skill with extracted frames and test cases.
