How to use the Openscreen CLI agent skill (cli-anything-openscreen)

What an agent skill is and what this one does

An agent skill is a folder containing a SKILL.md file that tells an AI agent when and how to use a set of tools. This skill's SKILL.md describes a stateful command-line interface for editing screen recordings. Once it is installed, the agent can run the cli-anything-openscreen commands for you.

The source lists these editing actions: zoom effects on chosen moments, speed changes for idle or important parts, trimming sections out, cropping the visible area, text annotations, background and padding settings, and exporting a final MP4 or GIF. The agent can also capture preview bundles (a low-res clip, sampled frames, a hero frame and a summary.json) so it can check its own edits.

Install the skill (ZIP or Claude Settings)

Download the skill ZIP from this page and unzip it into one of two places. For a personal setup, unzip it into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. The skill ships as a single file, SKILL.md.

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

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

Install the command-line tool

The skill drives a separate command-line tool, which you install with pip. You also need Python 3.10 or newer and ffmpeg installed on your system. The tool will not run without them.

After installing, you can check the commands the tool offers with the help flag.

• Python 3.10+
• ffmpeg installed on the system
• The pip package cli-anything-openscreen

Trigger the skill with natural requests

You do not need to type commands yourself. Describe the edit in plain words and name the recording file. Use absolute file paths, since the skill's guidance asks agents to do so. Good requests state the time range in your recording, the effect you want and the output format.

Times are handled in milliseconds by the tool, so a request like 'from 4 seconds to 9 seconds' is translated by the agent. Coordinates such as crop or focus points are normalized from 0 to 1, so you can describe them as fractions of the frame.

• Make a 16:9 demo from /Users/me/capture.mp4, trim the first 3 seconds, and export an MP4.
• Zoom to depth 3 between 5000 and 8000 ms, focused on the top-left quarter of the frame.
• Speed up the idle section from 20 to 40 seconds to 2x and export a GIF.

Example workflow from raw capture to export

A typical run starts by creating a project from the recording, adding the edits, then rendering. The agent can save the project as a .openscreen file so you can reopen it later. Before exporting, it can use preview capture to check zoom, trim, annotation and timing.

The commands below are the ones shown in the source. The agent chooses the exact values from your request.

• Create a project with the source video
• Add zoom, speed, trim, crop or annotation regions
• Capture a quick preview bundle and review the hero frame and summary.json
• Render the final file and probe it to confirm it exists

Safety, permissions and what to review

The skill runs scripts, so it executes the cli-anything-openscreen tool on your machine. According to the listing, it does not use the network. The files describe local editing only. The agent will still read and write video files on your machine, so review which paths it touches, especially for confidential recordings.

Keep recordings you need to protect in a folder the agent is allowed to access, and check the rendered output before sharing it.

• The skill runs scripts; read SKILL.md and the commands before approving runs.
• The agent reads and writes video files on your disk.
• Confidential recordings: review what the agent accesses.

Tips for reliable results

Ask the agent to use JSON output, since the source recommends machine-readable results for agents. Ask it to check return codes and read error messages if a command fails. After rendering, have it probe the output file to confirm it exists and matches what you expected.

Setting padding, wallpaper, aspect ratio and export format is handled through project settings, so you can ask for them in the same request as the edit.

• Ask for --json output and check the return code of each command.
• Use preview capture before the final export to catch timing mistakes.
• Pick an aspect ratio up front: 16:9, 9:16, 1:1, 4:3 or 4:5 are supported.
• Exports are mp4 or gif; quality can be medium, good or source.