alexeygrigorev/.agents/fetch-google-recorder

Fetch Google Recorder

Fetch Recorder-provided transcripts from public Google Recorder share links. Fetch audio only when explicitly needed, and optionally transcribe audio with WhisperX.

Requirements

Google Recorder links must be public. Private links redirect to Google sign-in and cannot be fetched without an authenticated browser session or cookies.

Fetch Recorder Transcript

Prefer this when the user asks for the transcript. Google Recorder shares already expose a transcript through the Recorder playback RPC, so there is no need to download the .m4a or run WhisperX.

Run from the directory where outputs should be written:

uv run python ~/.claude/skills/fetch-google-recorder/recorder.py transcript \
  'https://recorder.google.com/<share-id>' \
  --out-dir fetched

This writes:

<recording-title>.transcript.txt
<recording-title>.transcript.timestamped.txt
<recording-title>.transcription.jsonpb

Use .transcript.txt for normal text workflows. Use .transcript.timestamped.txt when you need timestamps (e.g. placing chapter markers); each line is prefixed with its start time:

[HH:MM:SS] transcript text

Fetch Audio

Run from the directory where outputs should be written:

uv run python ~/.claude/skills/fetch-google-recorder/recorder.py fetch \
  'https://recorder.google.com/<share-id>' \
  --out-dir fetched

This writes the original .m4a audio file. The filename is lowercase and underscore-separated, based on the Recorder title. It also tries to write the Recorder-provided transcript files.

Transcribe Audio

WhisperX currently needs Python 3.13 for the available Torch wheels. Use --python 3.13 even if the surrounding repo defaults to another Python version:

uv run --python 3.13 --with whisperx --with torch \
  python ~/.claude/skills/fetch-google-recorder/recorder.py transcribe \
  fetched/google_recorder_apr_24_at_12_26.m4a

This writes:

<audio-stem>.whisperx.txt

The timestamped transcript format is:

[MM:SS] transcript text

Add --raw only if a plain raw transcript is also needed:

<audio-stem>.whisperx.raw.txt

Fetch And Transcribe

For the full flow:

uv run --python 3.13 --with whisperx --with torch \
  python ~/.claude/skills/fetch-google-recorder/recorder.py fetch-transcribe \
  'https://recorder.google.com/<share-id>' \
  --out-dir fetched

Output Use

For transcript requests, use the transcript command first. It is faster and avoids downloading audio.

After fetching/transcribing audio, use the .m4a for audio workflows and .whisperx.txt for WhisperX text workflows. If WhisperX transcription quality is poor, inspect the Recorder-provided .transcript.txt.