| name | meeting-transcriber |
| description | Transcribe macOS meeting audio from the project folder with a bundled whisper.cpp CLI, downloading the selected model on first use, and write transcript files back to the project folder. |
| version | 0.1.1 |
| runtime | {"engine":"deno"} |
| permissions | {"tensorpm":{"action_items":"none","decisions":"none","history":"none"},"network":["huggingface.co:443","cdn-lfs.hf.co:443","cas-bridge.xethub.hf.co:443","transfer.xethub.hf.co:443"],"project_files":{"read":["meetings/**","skill-cache/meeting-transcriber/**","transcripts/**"],"write":["transcripts/**","skill-cache/meeting-transcriber/**"]},"run":["system:afconvert","skill:assets/bin/darwin-arm64/whisper-cli"],"ai_provider":"none"} |
| limits | {"cpu_seconds":300,"memory_mb":2048} |
| scripts | {"transcribe":{"entrypoint":"scripts/transcribe.ts","description":"Decode a project audio file with macOS afconvert, run the bundled whisper.cpp CLI, and save transcript artifacts under transcripts/.","schema":{"input":{"type":"object","properties":{"audio_path":{"type":"string","description":"Project-relative path under meetings/, e.g. meetings/client-sync.m4a."},"model":{"type":"string","enum":["tiny","base","small"],"description":"Whisper model to download/use from the local project cache. Defaults to base."},"language":{"type":"string","description":"Optional Whisper language code, e.g. de or en."}},"required":["audio_path"]},"output":{"type":"object","properties":{"audio_path":{"type":"string"},"model":{"type":"string"},"transcript_markdown_path":{"type":"string"},"transcript_text_path":{"type":"string"},"transcript_vtt_path":{"type":"string"},"transcript_json_path":{"type":"string"}},"required":["audio_path","model","transcript_markdown_path","transcript_text_path"]}}}} |
Meeting Transcriber
Use this skill when a user has placed a meeting audio file under meetings/
in the project folder and wants a local transcript written under
transcripts/.
This is intentionally macOS-only for the first version:
afconvert decodes the source audio to 16 kHz mono WAV.- A
whisper.cpp whisper-cli binary must be shipped in the installed skill
payload at assets/bin/darwin-arm64/whisper-cli.
- The selected ggml model is downloaded from Hugging Face on first use and
cached in
skill-cache/meeting-transcriber/models/.
The script returns project-relative paths for the generated Markdown, TXT,
VTT, and JSON outputs. It does not create TensorPM action items by itself;
the agent should read the transcript and use TensorPM's existing proposal or
action-item tools for human-reviewed meeting notes.
