How to use the quietshrink screen recording compressor skill

What does the quietshrink skill make the agent do?

An agent skill is a folder of instructions (here, a SKILL.md file) that tells an AI agent how to handle a specific kind of task. This skill covers compressing video files such as .mov and .mp4, mainly screen recordings, so they are small enough to send through chat, email or GitHub.

The skill uses the Media Engine, which is Apple Silicon's hardware HEVC encoder, not the CPU. The source describes encoding as fast and silent. Before compressing, the agent can also inspect a file's codec, resolution, frame rate and duration with the probe command.

• Compress a file with one of four quality presets: tiny, balanced, transparent (default) or pristine.
• Check a video's codec, width, height, frame rate and duration before processing.
• Check that the environment is set up correctly with the doctor command.
• Return machine-readable JSON output when requested with --json.

How do you install the quietshrink agent skill?

Download the skill ZIP from the agent-skills directory page, then unzip it into one of these locations. For all your projects, use ~/.claude/skills/. For a single project, use the .claude/skills/ folder inside that project.

If you use the Claude apps instead, open Settings → Capabilities → Skills and upload the skill there.

The skill also depends on the cli-anything-quietshrink command and the quietshrink tool, which are installed through the project's install script. The skill itself does not install them. Check the source repository for the install script's exact steps.

• Requirement: macOS on Apple Silicon (M1 to M4) is recommended.
• Requirement: ffmpeg with hevc_videotoolbox support, installable with Homebrew.
• The skill ships as a single file, SKILL.md.

How do you trigger the skill with a natural request?

You do not need to name the command. Describe what you want, and the agent will match the request to the skill's description. Good requests name the file and the goal, such as making a recording small enough to email.

If you want a specific quality level, say so. Otherwise the agent uses the default transparent preset. The source says the agent should run doctor before processing, and probe on unfamiliar files.

• Compress ~/Desktop/demo.mov so I can send it in Slack.
• This screen recording is too big for GitHub. Make it smaller without visible quality loss.
• Shrink this recording with the smallest quality you think is acceptable for chat.
• Check the codec and length of this video before you compress it.

Which quality preset should you pick?

The skill provides four presets. The figures below are typical reductions and SSIM scores listed in the SKILL.md file, not guarantees for your footage. SSIM is a similarity score where closer to 1 means the output looks more like the original.

Tiny is for chat and email where small artifacts are acceptable. Transparent is the default and is described as visually lossless, so use it for anything important. Pristine is for archiving or editing.

• tiny (q 50): typical reduction about 90%, SSIM about 0.95, for chat and email.
• balanced (q 55): typical reduction about 88%, SSIM about 0.99, for docs and sharing.
• transparent (q 60, default): typical reduction about 87%, SSIM about 0.99 or higher, for important files.
• pristine (q 70): typical reduction about 84%, SSIM about 0.997, for archival near-source quality.

What output does the skill return?

The compress command reports the input and output paths, file sizes in bytes, saved bytes and saved percent, duration, elapsed time, encoding speed, the preset used, the q value and the GOP setting. The probe command reports the path, size in bytes and megabytes, codec, width, height, framerate and duration.

If something fails, the source lists three common errors. ffmpeg not found is fixed with brew install ffmpeg. hevc_videotoolbox not available is fixed with brew reinstall ffmpeg. compression_failed usually means the input is damaged, and running with --verbose shows the ffmpeg errors.

• Run doctor first to confirm the environment.
• Use probe on unfamiliar files.
• Add --json to any command for machine-readable output.

What should you know about safety, scripts and network use?

This skill runs scripts, because it calls the cli-anything-quietshrink command and ffmpeg on your Mac. It does not use the network, according to the listing. Read the install script and the source repository before running it, as you would with any tool that runs on your computer.

The skill is not suited to every file. Camera footage and vlogs will shrink much less. The default preset is lossy, so if you need a byte-identical or lossless copy, keep the original. On Intel Macs and other systems, the skill falls back to less efficient software encoding.

• Keep your original recording until you have checked the compressed copy.
• Source repository: https://github.com/achiya-automation/quietshrink
• Background reading on the approach: WHY.md in the same repository.