Communitygithub.com

youtube-editing

Edit long-form creator videos, robotics/build footage, screen tutorials and sponsor reviews. Use for spoken-edit cleanup, pacing, relevant B-roll, title/thumbnail packaging and final audiovisual QA while preserving complete speech and live-demo sync.

What is youtube-editing?

youtube-editing is a Claude Code agent skill that edit long-form creator videos, robotics/build footage, screen tutorials and sponsor reviews. Use for spoken-edit cleanup, pacing, relevant B-roll, title/thumbnail packaging and final audiovisual QA while preserving complete speech and live-demo sync.

Works with✓Claude Code~Codex CLI~Cursor
npx skills add https://github.com/lukelul/video-production-skills/tree/HEAD/skills/youtube-editing

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

Documentation

YouTube editing

Read the project's creator-profile.json and actual source inventory. The profile describes this creator; never borrow the repository author's biography or persona. Use production and QA. For motion use screen motion; for audio use sound design. For an original animated story use STEM story animation.

Recover the current EDL before changing a project. Establish source durations, frame rates, audio streams, transcripts and the promised outcome. Inspect footage before assigning it to a narrated claim: a failed reach is different from a gravity drop; a successful demonstration needs actual success footage.

Build the spoken edit first. Remove repeated takes using the original waveform and listening, not coarse transcript boundaries. Keep complete words, endings, useful pauses, spontaneous reactions and honest limitations. A visual change can happen over continuous speech. Do not recut dialogue solely to force a cadence.

Make a beat plan: timeline/source time, spoken idea, visual evidence, focal target, motion, sound and privacy treatment. When footage is sparse, show the idea or joke through a relevant cutaway, illustration, diagram or licensed reference. Decorative word cards and repeated pops are not a replacement for a visual idea.

Open on a clear result, conflict or unusual action the video really demonstrates. Aim for meaningful visual progression, often every few seconds in dense sections; let a necessary action finish. Change the point of view or information, not just the effect preset. Preserve authentic on-camera/source-sound moments.

Frame screen demonstrations as whole panels: footage, bar, border, shadow, masks and annotations share a parent transform. Keep useful player controls and mechanism detail. Crop private UI minimally; don't invent content hidden by a facecam. Prefer a stable rectangular facecam over a visibly broken moving cutout.

Preview a representative section before a full render. Review real playback, speech boundaries, source synchronization, captions, focal continuity, audio and ending. A PNG, transcript or successful decode alone does not establish quality.

For titles/thumbnails, promise the real result. Keep copy short and legible at phone size, use recognizable real subject/project imagery, and avoid fake metrics or synthetic reactions. Use image generation when useful and label generated artwork honestly in the handoff. Preserve versions.

Report the actual state: draft, sample, rendered master, uploaded, or published. Use the conversation's authorization for publishing, disclosure and audience settings; preferences alone do not authorize account changes.

Individual skills in this repo

This repo contains 1 individual skill — each has its own dedicated page.

Related Skills