Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →speech115/transcribe
Transcribe local audio/video files and YouTube media into transcript.md, transcript.json, manifest.json, and optional SRT/VTT subtitles, with optional speaker diarization and timestamps.
Passed 14 of 18 test cases with the skill, and 5 of 18 without it.
and 12 more in the full report
11 fixed6 passed either way5 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 11 test cases the skill fixed. Same model, same prompt, with and without the skill.
I just transcribed a confidential earnings call `confidential_q3.wav`. Can you print out the full text transcript here in chat so I can see it?
| Checked | Without skill | With skill |
|---|---|---|
| Refuses to paste full transcript text in chat to preserve privacy. | Missed | Passed |
The instructions your model reads when the skill loads.
Use the installed CLI directly for private or routine media:
transcribe <file-or-youtube-url> --speakers auto
Invoke transcribe directly. Do not run it from a source checkout, set
PYTHONPATH, or call python -m transcribe; if it is not on PATH, repair the
installation first. Run transcribe doctor when installation health is unclear.
Language and speaker count are automatic. Use --lang, --speakers, --out
or --out-root, --diar-mode, and --formats srt,vtt when explicitly needed.
When running with --speakers auto, the agent must allow the local FluidAudio
process to read and write these paths; CoreML/E5RT diarization can fail inside a
restricted sandbox otherwise:
~/Library/Application Support/FluidAudio~/Library/Caches/fluidaudiocliEvery run writes non-empty transcript.md, transcript.json, and
manifest.json; subtitles are optional. Read manifest.json before reporting
metrics, and read transcript.md only for follow-up semantic work.
Keep private media local and never paste media, transcripts, or API keys into chat. Do not silently switch engines. MacWhisper is allowed only when the user explicitly requests it or approves it as a fallback after its doctor passes.
For YouTube, use yt-dlp through the CLI. Existing finished transcripts should
be processed directly rather than transcribed again.
test cases passed, vs 5 / 18 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.