The Agent Skills Directory

[COMMAND_EXECUTION]: The skill generates a single-purpose Node.js script from a local template (templates/record.mjs.template) and executes it to record browser interactions.
It employs strict sanitization by using JSON.stringify for all user-controllable variables (URLs, selectors, names) before they are injected into the template, effectively preventing JavaScript injection.
Interaction logic is restricted to a predefined catalog of 'recipes' or a structured 'multi' action list that is validated against a whitelist of allowed commands, preventing the execution of arbitrary user-supplied code.
[COMMAND_EXECUTION]: Uses ffmpeg and ffprobe for video processing tasks such as cropping, transcoding, and frame-rate analysis.
The bash script responsible for these operations (templates/crop.sh.template) performs validation on its inputs, ensuring that slugs are alphanumeric and bounding box coordinates are integers within the source video's dimensions.
[EXTERNAL_DOWNLOADS]: The skill identifies and checks for the existence of necessary tools like Playwright and ffmpeg during its preflight phase.
It follows best practices by not installing these tools silently; instead, it generates an installation plan and requires explicit user approval before proceeding with package manager commands.
[DATA_EXFILTRATION]: While the skill can handle local authentication states for browser sessions, it includes built-in safeguards to prevent data exposure.
It implements mandatory gating and warning messages when the user attempts to record a non-localhost URL or use an existing authentication state, ensuring the user is aware of potential PII or credential capture before the video is created.

screen-recorder