Editing a vertical short is often more granular labor than it looks: trimming micro-pauses, aligning screen recordings to narration, captioning, layering graphics and balancing music can turn a one-minute clip into dozens of small edits. In a hands-on test, Claude Code—using a free /shorts-editor skill plus FFmpeg, Whisper and HyperFrames—assembled a finished 70‑second Short with 98 clips across 13 tracks in roughly 40 minutes, with no manual timeline dragging.

What Claude Code can (and can’t) take off your plate

Claude Code doesn’t mimic a GUI editor like Premiere Pro. It treats editing as a programmatic task: transcribe, detect cut points, map assets to layouts and render a timeline via other tools. That means it can reliably:

  • Transcribe speech and attach timestamps for precise cuts.
  • Remove pauses and select the best takes when lines repeat.
  • Apply consistent captions and graphics based on reference videos.
  • Mix audio elements and output a vertical (1080×1920) render.

Practical limits matter: you need a developer-friendly environment and several third-party utilities. Claude Code won’t replace creative judgment—expect to review outputs, correct captions or tweak layouts, and refine the reference set that teaches its style.

What you need to run this workflow

The proof-of-concept combines a Claude Code skill with command-line tools that handle transcription, media processing and rendering. Core components:

  • Claude Code (with the /shorts-editor skill installed)
  • FFmpeg (media manipulation)
  • Whisper (speech-to-text for timestamps)
  • Node.js 18+ and HyperFrames (timeline builder and renderer)

Verify installations with the respective commands:

ffmpeg -version
whisper --help
node -v
npx hyperframes --version

A five-step repeatable editing workflow

The approach reduces short-form edits to five repeatable tasks. Build the skill and style guide once; reuse them for future clips.

1) Install the editing skill

Install the /shorts-editor skill so Claude Code can interpret assets, choose layouts and output a HyperFrames project. The skill encodes your editing logic and produces a renderable timeline.

2) Arrange a single project folder

Keep every asset together so the skill operates from a defined input set. A reusable structure looks like:

my-short/
  footage/
  broll/
  music/
  sfx/
  brand/
  references/

Put 3–10 finished videos that reflect your typical editing style into references/—Claude uses them to learn caption style, framing and pacing.

3) Teach Claude your style

Run the teach command so Claude analyzes the reference videos and outputs a layout guide. That guide specifies caption positions, typical element durations and preferred transitions. The clearer the guide, the closer the automated edits will match your brand.

4) Run the automated edit

From the project folder, launch Claude Code and trigger the edit. The skill will:

  • Transcribe talking-head audio and attach word-level timestamps.
  • Detect natural cut points (complete sentences, long pauses) and choose the best takes when repeats exist.
  • Assemble talking head, screen recordings and result cards into a HyperFrames timeline.
  • Render a vertical export and save the timeline and layout guide for iteration.

In the test, this produced a 70-second vertical short in about 40 minutes without any manual timeline edits.

5) Inspect and iterate

The process saves the MP4 plus timeline data and a layout guide you can refine. Check captions, alignment and audio balance. If the output drifts from your standard, update your references or tweak the layout guide and rerun the edit.

Practical tips to improve results

  • Segment screen recordings by discrete steps or topics—short, labeled captures are easier to match to narration than one long recording.
  • Limit music choices to one to three licensed tracks and keep consistent SFX so the model maps audio reliably across projects.
  • Use 3–10 accurate reference videos; the skill’s style guide inherits your casing, caption placement and pacing choices from them.
  • Run edits on a machine with adequate CPU and disk I/O—transcription and rendering are resource-intensive.

Who should try this, and what to watch next

This setup suits teams that publish frequent short-form content and can invest in an initial engineering setup to gain recurring time savings. It’s useful when brand consistency across many clips matters more than bespoke, frame-by-frame creative decisions.

Operational checklist: maintain a clear reference library, validate captions and music clearances, and insert human review steps into your publishing pipeline. The workflow removes the repetitive timeline work but doesn’t eliminate editorial oversight.

If you want to experiment: prepare a single project folder, pick three representative references, confirm your toolchain installs, run the skill and review the render. Use the saved layout guide and timeline files to refine future edits until the outputs meet your team’s standards.