Transcribe audio

Transcribes a local file on the user's machine with faster-whisper (MIT, by SYSTRAN). Nothing is sent to any cloud service; the only network access is downloading the Python packages and the model once.

All work goes through scripts/transcribe.py in this skill's directory. Do not install other transcription tools, do not search package indexes, and do not improvise alternative packages.

Steps

1. Resolve the input

• Get the file path from the user. Expand ~ and confirm the file exists.
• If the user asked for a specific language, note its code (en, pl, de, ...). Otherwise leave it to auto-detection.
• If the user asked for subtitles, use --format srt (or vtt). Default is txt.

2. Make sure uv is available

Run uv --version. uv runs the script in an isolated, cached environment with pinned dependencies, so nothing is installed globally.

If it is missing, ask the user before installing. Offer: