The Agent Skills Directory

[COMMAND_EXECUTION]: The skill leverages ffmpeg for sophisticated video processing, including silence detection, audio denoising (afftdn), and multi-pass trimming. It also utilizes node to execute local logic for transcription and sentence reconstruction.
[EXTERNAL_DOWNLOADS]: The skill documents that the first execution will download the Whisper machine learning model (approximately 1.5GB) for local transcription tasks. This is standard behavior for media processing tools and is clearly disclosed.
[DYNAMIC_EXECUTION]: The skill executes local TypeScript and CommonJS modules (./src/lib/transcribe.ts and scripts/sentence-from-captions.cjs) using the node runtime to handle transcription data and redo-take detection logic.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input in the form of user-provided video files and the resulting transcripts. It includes a security best practice by requiring explicit user confirmation via a markdown table before applying any automated redo cuts, ensuring the agent does not perform destructive edits without oversight.

short-form-editing