What problem does it solve? Analyzing long video or audio files manually is slow and impractical. This Skill automates the full pipeline: ingesting local media files, extracting keyframes and audio transcripts, running structured vision analysis through Gemini, answering natural-language questions about the content, and extracting clips around specific moments. ## Core Features & Use Cases - Media Ingestion & Preprocessing: Register video, audio, or image files with content-hash deduplication, then segment videos, extract downscaled keyframes, detect dead time, and optionally transcribe per-segment audio. - Structured Vision Analysis: Send keyframes or the full video to Gemini 2.5 Flash with a custom extraction prompt and JSON Schema, with concurrency pooling, retries, resumability, and cost tracking. - Natural-Language Q&A and Clip Extraction: Query the analysis results through Claude (e.g., "what happens at 3:14") and cut video clips with configurable pre/post-roll padding via ffmpeg. - Use Case: A coach uploads a two-hour game recording, runs a broad descriptive analysis pass, then asks questions like "find all the goals" and extracts highlight clips for each moment without reprocessing the video. ## Quick Start Ingest the video at /path/to/game.mp4, analyze it with a broad scene-description prompt, then ask what happens at the 15-minute mark and generate a clip of that moment.
