The Agent Skills Directory

[DYNAMIC_EXECUTION]: The template/src/kit.js file contains a call to eval() within the window.renderSheet function. This is used to evaluate strings passed to the at parameter, which allows for dynamic coordinate expressions (e.g., PLK.MX(1.38)). Because this input originates from the --crop-at command-line argument in render.mjs, it poses a risk of arbitrary JavaScript execution in the headless browser context if an attacker can control the CLI arguments.
[COMMAND_EXECUTION]: The template/render.mjs script executes external commands using node:child_process.spawn. Specifically, it calls ffmpeg to encode rendered frames into an MP4 video file. This is standard functionality for a video engine but involves interaction with the host operating system's shell.
[DYNAMIC_EXECUTION]: The template/src/whiteboard.js file uses synchronous XMLHttpRequest in the _loadText function to read the source code of all currently loaded scripts. This is done to identify character sets for font optimization but involves dynamic reading and processing of code files.
[INDIRECT_PROMPT_INJECTION]: The skill provides an animation engine that renders user-supplied text through functions like writeOn and wbCaptions. This constitutes an indirect prompt injection surface where untrusted data from the agent's context is processed and displayed, although the risk is limited to the visual rendering output.

whiteboard