The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Python subprocess module to orchestrate video processing tasks via external command-line tools.
Evidence: Multiple scripts including assets/tools/build.py, assets/tools/build_vo.py, assets/tools/pipeline.py, assets/tools/download.py, assets/tools/transcribe.py, assets/tools/transcribe_cloud.py, and assets/tools/batch_vo.py execute ffmpeg, ffprobe, and other local utilities.
[EXTERNAL_DOWNLOADS]: The skill allows the agent to fetch media files from external URLs to be used as source material for editing.
Evidence: assets/tools/download.py employs the standard yt-dlp utility to download video and audio content from various web platforms.
[DYNAMIC_EXECUTION]: The skill generates visual overlays by rendering HTML and CSS templates at runtime.
Evidence: assets/tools/render_overlays.py invokes a headless Chrome browser to convert user-defined or agent-generated HTML snippets into PNG images for video compositing.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data that could theoretically contain adversarial content, though it includes mitigations and warnings.
Ingestion points: The agent processes speech-to-text transcripts (transcript.json) and user-supplied metadata in overlays.json and edl.json.
Capability inventory: The skill can execute commands, render HTML, and write files based on processed input.
Sanitization: Basic character escaping is performed for subtitles, and the skill includes extensive troubleshooting documentation regarding model hallucinations.
Boundary markers: The skill instructions emphasize manual verification of transcripts and visual inspection of rendered frames to prevent errors or malicious injection.
[DATA_EXFILTRATION]: The skill communicates with external APIs to perform resource-intensive tasks like transcription.
Evidence: assets/tools/transcribe_cloud.py transmits audio data to well-known service providers (Groq and OpenAI) when local transcription is not available. These operations are transparent and consistent with the skill's documented functionality.

videoediting