Using the OpenAI Screenshot Skill safely

The Screenshot Skill is used as operating guidance inside Codex or a compatible harness. Its primary source is the screenshot directory in OpenAI's official curated skills repository.

Clarify the request and destination

First confirm that a capture is actually requested and whether a destination path was supplied. Preserve the path requested by the user exactly. Without a path, use the expected default location. For an internal visual check, a temporary location is preferable to the project directory.

Prefer an integrated tool

Use the capture function built into the relevant tool when the content comes from a browser, a Figma file, or an Electron application. Use the operating-system workflow as the fallback for an explicitly requested desktop capture or when no more targeted function is available.

Select the Linux helper

On Linux, the official helper selects the first available option from scrot, gnome-screenshot, and ImageMagick import. A full-screen capture covers the virtual desktop. For a region, use a region only when scrot or ImageMagick is installed. If no tool is available, the environment must be prepared first; do not record an invented success.

Review the image before sharing

Inspect the finished image for notifications, private material, file paths, customer data, and secrets. Crop unnecessary edges or repeat the capture with a smaller region. A screenshot containing a visible token must be treated as a possible secret exposure and must not be shared.

Approval and retention

Before publication or sharing, a person should review the purpose, recipient, and remaining information. Name the file clearly, retain it only as long as needed, and remove temporary review files from shared directories. The technical capture does not replace a privacy or release decision.