The Agent Skills Directory

[COMMAND_EXECUTION]: The render.mjs script utilizes the node:child_process module to spawn ffmpeg and google-chrome (via Puppeteer) for video encoding and frame rendering tasks. The new_project.sh script also executes shell commands like rsync, cp, sed, and npm to scaffold projects.
[DYNAMIC_EXECUTION]: The template/src/kit.js file contains a call to (0, eval)(v) within the window.renderSheet function. This is designed to evaluate string expressions for crop coordinates and world points, allowing for dynamic positioning during the rendering of contact sheets.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch external resources. The new_project.sh script runs npm install, downloading packages from the NPM registry. The studio.html file loads font families from fonts.googleapis.com. These are well-known and established services.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to generate animation code based on user-supplied briefs or reference descriptions. This creates an attack surface where a malicious brief could influence the agent to generate code that performs unintended actions when executed in the local rendering environment.
Ingestion points: User-provided animation briefs and reference video descriptions defined in the SKILL.md workflow.
Boundary markers: None present; the skill does not define delimiters or warnings to ignore embedded instructions in the generated scripts.
Capability inventory: The environment supports file system access, network downloads, dynamic JavaScript evaluation (eval), and subprocess spawning (ffmpeg, chrome).
Sanitization: The skill lacks explicit sanitization or validation logic for the code expressions passed to the rendering engine.

pixel-art