The Agent Skills Directory

[COMMAND_EXECUTION]
The Rust orchestration engine and several Node.js utility scripts are designed to execute external commands and project-specific scripts.
The engine specifically executes "project adapters" (such as render_segment.py, review_segment.py, and review_final.py) which are expected to be provided within the project directory. This poses a significant risk if project files originate from untrusted sources.
The engine spawns ffmpeg for video assembly and frame validation and executes Python scripts like analyze.py and fetch_assets.py.
The scripts/audio.mjs script facilitates the execution of an audio engine binary, which can be overridden by the HF_MEDIA_ENGINE environment variable.
[DYNAMIC_EXECUTION]
The rendering scripts skills/anime-cel/template/render.mjs and skills/crayon-storybook/template/render.mjs, as well as scripts/preview-frames.cjs, use eval() to parse coordinate expressions provided via command-line arguments.
This allows for arbitrary JavaScript execution within the headless browser instances used for rendering if the input to these scripts is not strictly validated.
[EXTERNAL_DOWNLOADS]
The skill's workflow includes downloading video content from arbitrary URLs using yt-dlp during the analysis phase as described in SKILL.md.
[INDIRECT_PROMPT_INJECTION]
The skill analyzes and transcribes external media, creating a vulnerability surface where malicious input content could influence the agent's behavior.
Ingestion points: External media files and URLs are processed in the analysis phase to generate report.json and STYLE.md (e.g., in skills/video-clone/scripts/analyze.py).
Boundary markers: No explicit delimiters are used to wrap content extracted from reference media before it enters the agent context.
Capability inventory: The skill has powerful capabilities, including file system read/write access and the ability to spawn multiple subprocesses (FFmpeg, Python, Node.js).
Sanitization: While the engine performs path traversal checks and rejects symlinks (in engine/src/project_paths.rs), it lacks sanitization for the textual data extracted from reference videos (e.g., transcripts) before using it in the storyboard planning process.

video-style-cloner