The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to download and execute an installation script for the 'uv' tool from astral.sh. This is a well-known developer tool from a recognized technology provider.
[REMOTE_CODE_EXECUTION]: The skill uses uvx to dynamically download and execute the rembg package from PyPI for background removal tasks. It also installs the puppeteer package which downloads a headless Chromium browser.
[COMMAND_EXECUTION]: The skill frequently executes external CLI tools including ffmpeg, node, and a specialized codex CLI to process media and generate images.
[INDIRECT_PROMPT_INJECTION]: The skill takes user-supplied "animation ideas" and interpolates them into generated animation.html and capture.mjs files. These files are then executed by a headless browser and Node.js respectively.
Ingestion points: User-provided animation idea text and photo input are used to derive file names, prompt content, and script variables.
Boundary markers: Absent; the skill does not explicitly instruct the agent to sanitize or delimit the user input before placing it into the code templates.
Capability inventory: The skill has broad capabilities including file system writes, subprocess execution (ffmpeg, node), and network access (via Puppeteer/browser).
Sanitization: No explicit sanitization or escaping of the user-provided idea is specified before it is embedded into the executable HTML/JS templates.
[METADATA_POISONING]: The skill references an NPM package @openai/codex for image generation. While the name suggests an official OpenAI tool, it appears to be a third-party utility. This naming convention can be deceptive to users regarding the provenance and official support of the dependency.

vox