agentsclimarketplace

Transcribe

Skill TserenTserenov/FMT-exocortex-template/.claude/skills/transcribe

Exocortex template: fork & deploy your AI-powered personal knowledge system with Claude Code

Install
npx -y skills add TserenTserenov/FMT-exocortex-template --skill transcribe

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

What its author says it does

Copied from the file, not written here

Transcribe audio/video files via MLX Whisper (Apple Silicon). Usage: /transcribe path/to/file.mp3

SKILL.md

2.4 KB, 557 tokens by cl100k_base, as published. Nobody here has run it

Транскрипция аудио/видео

Транскрипция через MLX Whisper на Apple Silicon. Работает локально, без облака.

Расположение

  • Модели: ~/.local/share/mlx-whisper/mlx_models/
  • Venv: ~/.local/share/mlx-whisper/.venv-whisper/
  • Модель: large-v3 (точная, ~3 ГБ). Единственная используемая модель

Инструкция для Claude

Шаг 1: Проверка venv

~/.local/share/mlx-whisper/.venv-whisper/bin/python -c "import mlx_whisper; print('ok')" 2>/dev/null

Если ошибка (сломан или отсутствует) — пересоздать:

rm -rf ~/.local/share/mlx-whisper/.venv-whisper
python3 -m venv ~/.local/share/mlx-whisper/.venv-whisper
~/.local/share/mlx-whisper/.venv-whisper/bin/pip install mlx-whisper

Шаг 2: Определить файл и модель

  • Аргумент скилла = путь к файлу. Если не указан — спросить пользователя.
  • Всегда использовать large-v3. Других моделей нет.

Шаг 3: Транскрипция

bash "$IWE_SCRIPTS/route-task.sh" --skill transcribe --args "<путь_к_файлу>"

Если язык не русский — пользователь укажет, или скрипт автоматически детектирует.

Шаг 4: Результат

  • Показать текст пользователю.
  • Если пользователь просит сохранить — записать в файл рядом с исходным: <имя_файла>.txt.
  • Для длинных файлов (>30 мин) предупредить, что может занять несколько минут.

Поддерживаемые форматы

mp3, m4a, wav, flac, ogg, mp4, mkv, webm — любые, которые поддерживает ffmpeg.

<!-- USER-SPACE --> <!-- /USER-SPACE -->

Gives 0 of the 12 instructions most video audio skills give in 557 tokens

Counted across 621 of the 795 authors here whose files we hold, read 2026-08-06

  • read individual rule files for detailed explanationsin 21 of 621, across 9 files
  • Use WAV PCM 16kHz mono audio formatin 13 of 621, across 4 files
  • render final videoin 13 of 621, across 6 files
  • use this skill when dealing with Remotion codein 11 of 621, across 4 files
  • save generated audio to a WAV filein 11 of 621, across 4 files
  • handle conversion errors gracefullyin 10 of 621, across 6 files
  • add captions to videos alwaysin 10 of 621, across 4 files
  • generate music from text descriptions using MusicGenin 9 of 621, across 2 files
  • do not skip pipeline layersin 9 of 621, across 3 files
  • do not make one tool do everythingin 9 of 621, across 3 files
  • never ask the user to paste their full API keyin 9 of 621, across 3 files
  • use azure document intelligence for complex pdfsin 9 of 621, across 4 files

Said here and by no other author read

  • Verify the virtual environment exists and works
  • Recreate the virtual environment if it is missing or broken
  • Use the large-v3 model exclusively
  • Show the transcribed text to the user
  • Warn about processing time for files longer than 30 minutes

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.