What problem does it solve? Running media operations like transcription, segmentation, video transforms, and foreground matting against a backend job API normally requires manual upload handling, bearer-token management, and job polling. This Skill wraps all of that into one deterministic command surface that accepts a local file or URL and returns a stable JSON result envelope. ## Core Features & Use Cases - Artifact-backed transcription: Transcribe local or remote audio/video through the WIN local_transcription provider with diarization, caching, and optional speaker identification, returning transcript text plus transcript/words/sentences artifact URLs. - SAM 3.1 segmentation and matting: Segment images or videos by prompt with anchor/window controls, generating mask URLs by default and optional alpha artifacts, or produce foreground matte manifests. - Upload, transform, and job inspection: Upload local files to S3-backed storage, submit trim/scale/crop transform jobs, and poll or inspect existing jobs by job_id. - Use Case: Given a local podcast recording, run the transcribe subcommand to get the transcript text and cached artifact URLs written to a JSON file for downstream editing steps. ## Quick Start Ask the agent to transcribe a local audio file with the media-toolkit transcribe command and write the JSON result envelope to a file.
