The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes external commands to handle video and audio processing.
In scripts/beat_grid.py, subprocess.run is used to invoke ffmpeg for spectral analysis of audio files.
In template/render.mjs, spawn is used to call ffmpeg for encoding image sequences into MP4 videos and node to manage parallel worker processes.
These implementations use argument lists rather than shell strings, which is a safe practice that prevents shell injection attacks.
[DYNAMIC_EXECUTION]: The animation engine uses eval() in template/src/core.js within the headless Chrome environment. This is used to evaluate mathematical or helper-function expressions passed via the --crop-at command-line argument for camera tracking and review. Since this occurs in a sandboxed browser context driven by the user's agent, the security risk is minimal and serves a functional purpose.
[EXTERNAL_DOWNLOADS]: The skill manages several external dependencies and resources:
The scripts/new_project.sh script performs an npm install to download p5, p5.brush, and puppeteer-core from the official NPM registry.
The template/studio.html file loads fonts from fonts.googleapis.com (Google Fonts), a well-known and trusted service.
The scripts/beat_grid.py documentation mentions a dependency on numpy via pip.
[INDIRECT_PROMPT_INJECTION]: The skill handles untrusted external data which could serve as an attack surface for indirect prompt injection:
Ingestion points: Song lyrics (text), LRC files (timings), and audio files provided by the user or fetched by the agent.
Boundary markers: None explicitly defined in the prompts; the data is interpolated directly into JavaScript scene files (src/karaoke.js).
Capability inventory: The skill can write files to the local system, execute ffmpeg commands, and control a headless Chrome instance.
Sanitization: There is no explicit sanitization of lyric content before interpolation into the renderer's JavaScript context. However, the risk is mitigated by the fact that the resulting code is executed in a local development environment and typically reviewed by the agent before rendering.
[FILE_SYSTEM_ACCESS]: The renderer (render.mjs) manages extensive file system operations, including creating project directories, writing temporary JPEG/PNG frames, and outputting final MP4 videos. These operations are restricted to the project directory and the user-specified output paths.

painted-animation