The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes ffmpeg and ffprobe using Python's subprocess module to perform video editing tasks such as extraction, concatenation, and rendering. These operations are conducted through argument lists, which avoids shell injection risks, and include validation for frame rates and sanitization for paths.
[EXTERNAL_DOWNLOADS]: The skill integrates with well-known external services and libraries to fulfill its purpose:
Transcription is performed via the official ElevenLabs API.
Animations are created using established frameworks like Manim (programmatic math visuals), Remotion (React-based video), and HyperFrames (web-native video).
Standard utilities like yt-dlp are used for media retrieval.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted audio/video content and processes the resulting transcripts to drive editing decisions (via a generated Edit Decision List). This architecture creates a surface for potential injection via the processed media content. The risk is mitigated by using structured JSON schemas for logic, deterministic helper scripts for execution, and a multi-step confirmation process requiring user approval before destructive actions.

video-use