The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/audio.mjs generates a Python code block as a string and executes it using python3 -c. This generated script utilizes the transformers library to run MusicGen models for background music generation. The use of runtime-generated code is a pattern that allows for dynamic behavior based on interpolated variables, such as the BGM prompt.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process arbitrary text (articles, notes, topics) provided by the user to generate video content. Ingestion points: Untrusted data enters the pipeline through capture/extracted/visible-text.txt, which is read by the scriptwriting.md agent and used to drive subsequent automated steps. Boundary markers: The agent prompts in the agents/ directory utilize Markdown structure and structured blocks to organize instructions and data, although specific 'ignore embedded instructions' warnings are not prominently used for the input text. Capability inventory: The pipeline possesses extensive capabilities, including executing arbitrary shell commands via Bash (npx, ffmpeg, pip), making network requests to external AI service APIs (HeyGen, ElevenLabs, Google), and writing to the local filesystem and temporary directories. Sanitization: Basic sanitization logic exists in scripts/audio.mjs to strip specific XML-like tags, but the system primarily relies on the model's instruction following to handle user-supplied narrative content safely.
[EXTERNAL_DOWNLOADS]: The skill automates the installation of various software dependencies and tools at runtime if they are not detected on the system. It uses pip install to fetch Python libraries including transformers, torch, soundfile, numpy, google-genai, python-dotenv, and elevenlabs. It uses npm install for the puppeteer-core module. It uses npx hyperframes commands to initialize project structures and add components from the registry. All identified packages and tools are well-known technology services and established open-source libraries.
[COMMAND_EXECUTION]: The skill extensively utilizes system commands for media processing (ffmpeg, ffprobe), package management (pip, npm), and orchestrating the video build process through various Node.js scripts.

faceless-explainer