The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of several standard libraries and tools from trusted registries (NPM, PyPI). These include google-genai (Google), librosa, mlx-whisper, p5, and puppeteer-core.- [COMMAND_EXECUTION]: The skill executes local system commands (node, ffmpeg, uv, gcloud) to manage the development environment, process audio/video assets, and interface with Google Cloud services.- [DYNAMIC_EXECUTION]: Video frames are rendered by executing JavaScript code within a headless Chrome instance controlled by Puppeteer. This involves running painting logic (using p5.brush) defined in various script files.- [INDIRECT_PROMPT_INJECTION]: The skill employs a multi-agent architecture where the primary agent dispatches tasks to sub-agents based on content from STORYBOARD.md. While the prompt template includes boundary instructions (e.g., 'Do not run any git commands', 'Do not edit shared files'), this structure creates an attack surface for indirect prompt injection if the source storyboard or lyrics were to contain malicious instructions for sub-agents.- [PRIVILEGE_ESCALATION]: The documented publishing workflow describes the use of AppleScript and browser automation to upload videos to YouTube. This high-privilege activity is explicitly labeled as requiring human oversight and manual confirmation.

painted-music-video