YouTube Shorts

Mission

Provide a platform-focused preset over Short-form Repurposing: long-form source in, ranked vertical short candidates out. Keep the original source immutable, preserve exact timestamps and context, and make the difference between an editorial score and actual audience performance explicit.

Required inputs

• authorized source video URL or local file;
• target channel, language, caption style, and requested clip count;
• desired duration range or permission to use provider defaults;
• whether the output should be finished clips or timestamps for another renderer; and
• source rights, speaker consent, music rights, review owner, and budget.

If no platform is specified, use a 9:16 draft as an assumption and state it. Do not claim that a duration, hook, crop, or caption style is currently YouTube-compliant without a supplied or verified rule set.

Pipeline

1. Inspect source metadata when available and upload local footage. Record source ID, duration, dimensions, audio/language, provenance, and rights.
2. Invoke media.clip_video to propose highlights. Capture every candidate's start/end time, transcript or hook, editorial rationale, provider score if returned, and source parent.
3. Remove overlapping candidates and rank for standalone meaning, complete thoughts, hook clarity, payoff, audio quality, and visual legibility. A score is not a prediction of views or revenue. Return fewer clips when the source does not contain enough strong moments.
4. Reframe selected time ranges with media.reframe_video or the live autocrop schema. Default to 9:16 only when that assumption is accepted. Inspect speaker/product tracking, cut points, safe areas, and context after the crop.
5. Generate captions after the segment is selected. Proofread names, numbers, claims, punctuation, timing, contrast, and line length. Flag uncertain transcription rather than silently correcting meaning.
6. Return a ranked package with timestamps, titles/hooks, source links, output URLs, and QA. Publishing or scheduling is handled by a separate approved social agent.

The current utility shapes are documented in MODELS.md and muapi-video-tools.md: clipping can return highlight coordinates, autocrop takes explicit start/end seconds and a ratio, and captions are a separate operation. Do not assume one call performs all three stages.

Review the finished candidates with video QA, especially crop safety, caption timing, source context, audio, and synthetic media disclosure.

Output contract

Return rank, clip ID, source ID, exact timestamps, duration, hook/title, editorial rationale, score provenance, crop ratio, caption status, model/tool, prediction IDs, output URL, rights state, and QA findings. Save a manifest or receipt under .video/ only when permitted, using reports-and-receipts.md.

Never pad the requested count with weak moments, remove qualifying context, or present generated captions/crops as final without review.