video-editing-skill

Edit videos with natural language — trim, jump cut, caption, overlay, and speed up — all from your terminal.

---

Trim clips • Remove silence • Burn captions • Add text overlays • Adjust speed • Chain it all

Zero runtime dependencies beyond FFmpeg and Whisper. Pure Bash, composable, pipeline-ready.

Requirements • Quick Start • Capabilities • Pipeline • Scripts • License

---

Why This Skill?

AI agents are great at writing code — but when you ask them to edit a video, they reach for bloated Python libraries, spin up runtimes, and still can't chain operations together.

• Pure Bash + FFmpeg — no Python runtimes, no package managers, no build steps
• Composable pipeline — chain trim, jump cut, speed, caption, and overlay in a single command
• 3 caption styles — Hormozi, standard, and minimal — burned directly into the video
• Smart silence removal — auto-detects dead air with configurable thresholds
• Whisper transcription — generates SRT files locally, no API calls, no cloud dependency
• Multi-agent compatible — built for OpenClaw, Claude Code, and Codex; used and tested with OpenClaw
• Works standalone — every script runs independently, with or without an AI agent

---

Table of Contents

• Requirements
• Quick Start
• Capabilities
• Full Pipeline
• Scripts Reference
• Configuration
• Troubleshooting
• Project Structure
• Agent Compatibility
• Contributing
• License

---

Requirements

You must have the following installed on your machine before using this skill.

Dependency	Required	Purpose	Install
FFmpeg	Yes	All video processing (trim, jump cut, overlay, speed, caption burning)	See below
ffprobe	Yes	Media analysis (ships with FFmpeg)	Included with FFmpeg
Whisper	For captions	Audio transcription to SRT subtitle files	See below
Bash	Yes	Script runtime	Pre-installed on macOS/Linux
Python 3	For captions	Required by Whisper	Pre-installed on most systems

Install FFmpeg

# macOS (Homebrew)
brew install ffmpeg

# Ubuntu / Debian
sudo apt install ffmpeg

# Arch Linux
sudo pacman -S ffmpeg

# Windows (Chocolatey)
choco install ffmpeg

Install Whisper (optional — only needed for captions)

pip install openai-whisper

Whisper runs locally on your machine. No API keys, no cloud calls, no data leaves your computer.

Verify Installation

ffmpeg -version     # should print version info
ffprobe -version    # should print version info
whisper --help      # should print usage (only if you need captions)

Or run the onboard script to check everything at once:

./scripts/onboard.sh --check

---

Quick Start

Install the Skill

The onboard script checks dependencies, makes scripts executable, and registers the skill with OpenClaw and Claude Code:

./scripts/onboard.sh

Onboard options

Flag	Description
(no flag)	Default — symlinks skill to ~/.openclaw/skills/ and ~/.claude/skills/
--copy	Copy files instead of symlinking
--check	Check dependencies only, don't install
--uninstall	Remove the skill from all locations

Run Your First Edit

# Remove silence from a video
bash scripts/jumpcut.sh ~/video.mp4

# Trim to a specific range
bash scripts/trim.sh ~/video.mp4 --start 00:01:00 --end 00:05:00

# Full pipeline: trim, remove silence, add captions, speed up
bash scripts/edit.sh ~/video.mp4 \
  --trim-start 00:00:10 --trim-end 00:10:00 \
  --jumpcut \
  --caption --caption-style hormozi \
  --speed 1.25 \
  --output final.mp4

That's it. No install step, no config files, no build process.

---

Capabilities

Trimming

Cut video to specific start/end timestamps or duration. Uses fast codec copying — no re-encoding.

scripts/trim.sh video.mp4 --start 00:01:30 --end 00:05:00

Option	Description	Default
--start	Start timestamp (HH:MM:SS or seconds)	00:00:00
--end	End timestamp	End of file
--output	Output path	{name}_trimmed.{ext}

Jump Cuts (Silence Removal)

Auto-detects and removes silent sections. Reports time saved and percentage removed.

scripts/jumpcut.sh video.mp4 --threshold -30 --duration 0.5 --padding 0.1

Option	Description	Default
--threshold	Silence threshold in dB	-30
--duration	Min silence duration to cut (seconds)	0.5
--padding	Padding around speech (seconds)	0.1
--output	Output path	{name}_jumpcut.{ext}

Captions

Two-step process: transcribe audio with Whisper, then burn subtitles into the video.

Step 1 — Transcribe:

scripts/transcribe.sh video.mp4 --model base --language en

Step 2 — Burn captions:

scripts/caption.sh video.mp4 video.srt --style hormozi

Caption Styles

Style	Description
hormozi	Bold, centered, large — Alex Hormozi word-by-word impact style
standard	Traditional bottom subtitles with semi-transparent background
minimal	Small lower-third captions, clean and unobtrusive

Whisper Model Options

Model	Speed	Accuracy	Best For
tiny	Fastest	Low	Quick drafts, testing
base	Fast	Good	Most videos (default)
small	Medium	Better	Noisy audio
medium	Slow	High	Long-form content
large	Slowest	Highest	Maximum accuracy

Text Overlays

Add positioned text at specific timestamps with configurable appearance.

scripts/overlay-text.sh video.mp4 \
  --text "Subscribe!" \
  --start 00:01:00 --end 00:01:05 \
  --position bottom-right \
  --fontsize 48 --color white

Option	Description	Default
--text	Text to display	(required)
--position	Placement (see below)	center
--start	Display start time	00:00:00
--end	Display end time	End of file
--fontsize	Font size in pixels	48
--color	Font color	white
--bg-color	Background color with opacity	(none)

Positions: center • top • bottom • top-left • top-right • bottom-left • bottom-right

Speed Changes

Adjust playback speed with pitch-corrected audio. Handles extreme values via automatic filter chaining.

scripts/edit.sh video.mp4 --speed 1.5

---

Full Pipeline

The edit.sh orchestrator chains multiple operations in a single command. Operations execute in this order:

trim → jump cut → speed → caption → overlay

scripts/edit.sh video.mp4 \
  --trim-start 00:00:10 --trim-end 00:10:00 \
  --jumpcut \
  --jumpcut-threshold -25 \
  --caption --caption-style hormozi \
  --speed 1.25 \
  --overlay-text "Like & Subscribe" \
  --overlay-start 00:01:00 --overlay-end 00:01:05 \
  --output final.mp4

All Pipeline Options

Option	Description
--trim-start	Trim start timestamp
--trim-end	Trim end timestamp
--jumpcut	Enable silence removal
--jumpcut-threshold	Silence threshold (dB)
--jumpcut-duration	Min silence duration (seconds)
--jumpcut-padding	Padding around speech (seconds)
--speed	Playback speed multiplier
--caption	Enable captioning (auto-transcribes)
--caption-style	Caption style: hormozi, standard, minimal
--caption-srt	Use existing SRT file instead of transcribing
--caption-model	Whisper model for transcription
--caption-language	Language code for transcription
--overlay-text	Text overlay content
--overlay-start	Overlay start timestamp
--overlay-end	Overlay end timestamp
--overlay-position	Overlay position
--overlay-fontsize	Overlay font size
--overlay-color	Overlay font color
--output	Final output path

Intermediate files are automatically cleaned up. The pipeline reports output file path, duration, and file size on completion.

---

Scripts Reference

Script	Purpose	Output
scripts/edit.sh	Main orchestrator — chains all operations	{name}_edited.{ext}
scripts/trim.sh	Trim by timestamps	{name}_trimmed.{ext}
scripts/jumpcut.sh	Remove silence via silencedetect	{name}_jumpcut.{ext}
scripts/transcribe.sh	Whisper transcription to SRT	{name}.srt
scripts/caption.sh	Burn SRT captions with style	{name}_captioned.{ext}
scripts/overlay-text.sh	Add positioned text overlays	{name}_overlay.{ext}

All scripts accept --help and include usage documentation in their headers.

---

Configuration

Environment Variable	Required	Default	Description
WHISPER_BIN	No	Auto-detected	Path to whisper binary

No config files, no JSON manifests. Every option is a command-line flag with sensible defaults.

---

Troubleshooting

"ffmpeg: command not found"

Install FFmpeg for your platform:

# macOS
brew install ffmpeg

# Ubuntu/Debian
sudo apt install ffmpeg

# Verify
ffmpeg -version

"whisper not found"

Install OpenAI Whisper:

pip install openai-whisper

# Or set a custom path
export WHISPER_BIN=/path/to/whisper

Jump cut removes too much / too little

Tune the silence detection parameters:

# More aggressive (removes more silence)
scripts/jumpcut.sh video.mp4 --threshold -25 --duration 0.3

# More conservative (keeps more pauses)
scripts/jumpcut.sh video.mp4 --threshold -35 --duration 1.0 --padding 0.3

Captions are inaccurate

Use a larger Whisper model for better accuracy:

scripts/transcribe.sh video.mp4 --model medium
# or for maximum accuracy:
scripts/transcribe.sh video.mp4 --model large

You can also specify the language explicitly:

scripts/transcribe.sh video.mp4 --model medium --language en

---

Project Structure

video-editing-skill/
  SKILL.md                  # Skill metadata for AI agents
  README.md                 # This file
  LICENSE                   # MIT License
  scripts/
    onboard.sh              # Setup: dependency check + skill install
    edit.sh                 # Pipeline orchestrator
    trim.sh                 # Video trimming
    jumpcut.sh              # Silence removal
    transcribe.sh           # Whisper transcription
    caption.sh              # Caption burning (3 styles)
    overlay-text.sh         # Text overlay insertion

---

Agent Compatibility

Built for OpenClaw, Claude Code, and Codex. Used and tested with OpenClaw.

This skill follows the Agent Skills Protocol via SKILL.md and works with any AI agent that can execute shell commands, including:

OpenClaw • Claude Code • Codex • Cursor • Windsurf • Cline • Roo Code • Goose • Continue • Kilo • Amp • and more

Natural language examples:

Edit my video at ~/video.mp4 — remove silence, add Hormozi-style captions, and speed it up 1.25x

Trim ~/interview.mp4 from 00:02:00 to 00:15:00 and add standard captions

Add a "Subscribe!" text overlay at the 1 minute mark in ~/video.mp4

The agent reads SKILL.md, identifies the operations needed, and calls the appropriate scripts.

---

Contributing

Contributions welcome! Please:

1. Fork the repo
2. Create a feature branch (git checkout -b feat/my-feature)
3. Test your changes against real video files
4. Submit a PR

---

License

MIT