메뉴
Use when user provides an audio file path, asks to transcribe audio, listen to a recording, or process voice messages. Triggers on file extensions like .opus, .mp3, .m4a, .wav, .ogg, .flac, .webm, or mentions of WhatsApp audio, voice notes, recordings, or podcasts.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | audio-transcriber |
| description | Use when user provides an audio file path, asks to transcribe audio, listen to a recording, or process voice messages. Triggers on file extensions like .opus, .mp3, .m4a, .wav, .ogg, .flac, .webm, or mentions of WhatsApp audio, voice notes, recordings, or podcasts. |
Transcribe audio files locally using OpenAI Whisper. Claude cannot hear — this skill bridges that gap by running Whisper on the user's machine.
.opus, .mp3, .m4a, .wav, .ogg, .flac, .webm, .aac, .wma)Before transcribing, verify the toolchain:
# 1. Check if Whisper is installed
which whisper
# 2. If not installed, install via pipx (preferred on macOS)
pipx install openai-whisper
# 3. If pipx unavailable, use pip with venv
python3 -m venv /tmp/whisper-env && source /tmp/whisper-env/bin/activate && pip install openai-whisper
whisper "<audio-file-path>" \
--model medium \
--output_dir "<scratchpad-dir>" \
--output_format txt
Use the session scratchpad directory for output — never pollute the user's project.
| Model | Size | Speed | Accuracy | Use When |
|---|---|---|---|---|
tiny | 39MB | Fastest | Low | Quick gist, short clear audio |
base | 74MB | Fast | Fair | Simple, clear speech |
small | 244MB | Moderate | Good | Most casual use |
medium | 1.5GB | Slow | Great | Default — best tradeoff |
large-v3 | 3GB | Slowest | Best | Non-English, heavy accents, noisy audio |
Default to medium. Upgrade to large-v3 if user reports poor quality or audio is non-English with heavy dialect.
--language de (ISO code or full name)--task translate| Format | Flag | Use Case |
|---|---|---|
.txt | --output_format txt | Default — plain transcript |
.srt | --output_format srt | Subtitles with timestamps |
.vtt | --output_format vtt | Web subtitles (WebVTT) |
.json | --output_format json | Programmatic access, word-level timestamps |
| all | --output_format all | Generate every format |
ls -la "<path>"medium model, output to scratchpad.txt output and present to user| Problem | Fix |
|---|---|
pip install blocked by PEP 668 | Use pipx install openai-whisper instead |
| Whisper not found after install | Check ~/.local/bin is in PATH, or use full path |
| Poor transcription quality | Upgrade to large-v3, or force correct --language |
| Unsupported format error | Convert first: ffmpeg -i input.xyz output.wav |
| Very long audio (>30 min) | Still works, just takes time. Warn user about duration. |
Whisper accepts multiple files:
whisper file1.opus file2.mp3 file3.m4a --model medium --output_dir "<scratchpad>"
Or loop for separate handling:
for f in /path/to/audios/*.opus; do
whisper "$f" --model medium --output_dir "<scratchpad>" --output_format txt
done
Build interactive canvas-based animations with pixel-art typography, spring physics, mouse/touch interaction, and idle animations. Use when asked to create animated backgrounds, letter grids, particle effects, or any interactive canvas component for React/Next.js sites.
Use when user asks to find, search, look up, or send iMessages, texts, or SMS. Triggers on mentions of iMessage, texts, messages, "what did X text me", "find my conversation with", "send a message to", "text X", or anything involving searching or sending iMessage/SMS/RCS messages.
Ganztagshelden Slack workspace communication rules. Use when composing or sending Slack messages to ensure proper thread structure, title formatting, reactions, and channel etiquette. Apply before sending any message via the Slack MCP.
Retrieves and queries up-to-date documentation and code examples from Context7 for any programming library or framework. Use when writing code that depends on external packages, verifying API signatures, looking up usage patterns, generating code with specific libraries, or when training data may be outdated. Covers setup questions, migration guides, and version-specific docs.
Use when user asks to find, search, or look up WhatsApp messages, chats, or conversations. Triggers on mentions of WhatsApp, messages, chats, "what did X say", "what did X tell me", or anything involving searching WhatsApp message history.
Copy text to the system clipboard. Use when user asks to copy something to clipboard, copy to pasteboard, or wants to paste something.