How to use the AI Video Clipping Expert agent skill

What an agent skill is and what this one does

An agent skill is a folder of instructions that an AI agent reads when a task matches it. This skill is documentation only: its file SKILL.md contains the commands, flags and steps for video clipping, and it does not run anything by itself. The agent uses that knowledge to carry out the work with the tools installed on your computer.

Once installed, the agent knows the specific commands for each stage of the job instead of guessing them. The stages are downloading the source video, extracting audio, transcribing to word-level timestamps, building subtitle files, finding scene cuts and silences, extracting clips, reformatting to 9:16 vertical, and burning captions into the output.

Install the clip-hand-skill on Claude

Download the ZIP for the skill from the page you are reading. The package contains two files: HAND.toml and SKILL.md.

Option A: put it in your personal skills folder by unzipping it into ~/.claude/skills/. Option B: put it in a single project by unzipping it into that project's .claude/skills/ folder.

Option C: in the Claude apps, open Settings, go to Capabilities, and upload the skill under Skills.

• Personal install: ~/.claude/skills/
• Project install: the .claude/skills/ folder inside your project
• Claude apps: Settings → Capabilities → Skills

Install the command-line tools the skill relies on

The skill assumes that the tools it calls are already installed and working. You need ffmpeg and ffprobe, yt-dlp, and OpenAI Whisper, or one of the alternatives named in the skill: whisper-ctranslate2 or mlx-whisper. The skill notes that mlx-whisper is a native option on macOS Apple Silicon.

Check that each tool runs from your terminal before you ask the agent to clip anything. If a command is not found, the agent cannot complete that stage.

• ffmpeg and ffprobe (video processing)
• yt-dlp (downloading video, audio and subtitles)
• OpenAI Whisper, or whisper-ctranslate2 or mlx-whisper (transcription)
• Optional: browser cookies, only for age-restricted videos

Trigger the skill with natural requests

You do not need special syntax. Describe the job in plain language and mention the source video and the result you want. The agent should pick up the skill when your request involves downloading, transcribing, captioning or cutting video.

Be specific about the source, the length of clips you want and the format. The more detail you give, the fewer questions the agent needs to ask.

• "Make three 30-second vertical clips with captions from this YouTube interview: [URL]"
• "Transcribe this recording and give me an SRT file for the first 10 minutes."
• "Find the scene changes and silences in this webinar so I can pick cut points."
• "Turn this 16:9 talk into a 9:16 clip with white captions burned in at the bottom."

Know the main techniques the skill teaches

Downloading: yt-dlp can fetch a video at up to 1080p or 720p, or extract audio only as WAV. It can also download manual or automatic subtitles and list the available subtitle languages.

Transcription: Whisper runs with word-level timestamps. The skill lists model sizes from tiny to large-v3 and recommends small as the default balance of speed and quality. Words with a probability below 0.5 are flagged as likely wrong.

Captions: the skill explains how to group words into lines of about 8 to 12 words, keep lines under 42 characters, and convert times so each clip starts at zero. Styled captions use ASS format.

Editing: ffmpeg is used to detect scene changes and silences, extract clips (re-encoded for accurate cuts or copied for speed), crop or pad to 1080x1920, and burn in subtitles.

Safety notes: scripts and network access

This skill is marked as running scripts and using the network. The tools it calls download content from the internet and process files on your machine. Review the commands the agent proposes before you approve them, especially any that download a URL or use browser cookies.

Only download videos you have the rights to use. The yt-dlp guidance can fetch arbitrary URLs and can use your browser cookies for age-restricted content. Doing this may breach platform terms or copyright law.

• Read each download URL before the agent runs it.
• Use the cookies option only when you need an age-restricted video and understand the terms involved.
• Keep your source files and output in a folder you control.

Tips for better clips

Start with a 720p download and the small Whisper model to test your pipeline quickly. Move to 1080p and medium or large-v3 for final output where accuracy matters.

Check the transcript before you trust the captions. Low-confidence words are the most likely errors. For vertical video, the skill offers two options: a center crop when the source is 16:9, or a scaled version with black letterbox padding that keeps the full frame.

• Use --restrict-filenames for safe filenames, as the skill recommends.
• Use forward slashes in ffmpeg subtitle paths, and escape the colon for absolute Windows paths.
• Prefer re-encoded extraction when cut points must be exact.