메뉴
Use when needing to process exported WhatsApp conversations — converts raw exports (zip or folder with _chat.txt) into clean markdown with transcribed voice messages
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
Use when needing to automate browser interactions, navigate websites, fill forms, take screenshots, extract data from web pages, or manage browser sessions via Playwright
Use when someone wants to capture their personal writing voice/style so Claude can write as them — building a "writing DNA" from their past emails, chat, posts, and docs across multiple media, and generating a portable per-person voice skill. Also use to refresh or upgrade an existing voice profile.
Find and deduplicate files in a project — replaces copies with symlinks, respects .gitignore and common exclusions. Use this skill whenever the user mentions duplicate files, file deduplication, saving disk space by removing copies, finding identical files, or cleaning up repeated documents in a repo. Especially relevant for documentation-heavy repos with PDFs or other large binary files that may have been copied around.
Use when needing to search, read, draft, send, download attachments from Gmail, or when any other skill requires Gmail operations
Use when needing to extract text, tables, or structured data from PDF files, as a fallback when the built-in Read tool cannot handle the PDF
Create clean, atomic commits following project conventions. Use when committing code changes, staging files, or writing commit messages.
| name | read-whatsapp-export |
| description | Use when needing to process exported WhatsApp conversations — converts raw exports (zip or folder with _chat.txt) into clean markdown with transcribed voice messages |
Converts raw WhatsApp exports into clean, chunked markdown files ready for reading. Fully offline.
process-conversation.py. Safe to re-run: it extracts into exported/, regenerates all chunks. A newer export just overwrites and regenerates.transcribe-conversation-audios.py (see below).process-conversation.py again to regenerate chunks with transcriptions inline.python3 ${CLAUDE_SKILL_DIR}/scripts/process-conversation.py "path/to/export.zip" --output-dir .
Always pass --output-dir . to create the conversation folder in the current project directory.
Options: --output-dir DIR (default: same parent as source), --max-chunk-kb N (default: 100).
Re-running with a newer export of the same conversation overwrites exported/ and regenerates all chunks. Existing transcriptions (.txt files) are preserved if the audio files haven't changed.
python3 ${CLAUDE_SKILL_DIR}/scripts/transcribe-conversation-audios.py "path/to/whatsapp-name-conversation" <language>
Transcribes all .opus files in exported/ that don't have a .txt yet. Saves the language and transcription preference in .processed. After transcription, re-run process-conversation.py to regenerate chunks with inline transcriptions. On future re-runs, process-conversation.py auto-transcribes new audio files if .processed has transcribe: true.
Language detection: Read the first ~50 lines of the conversation to identify the language (<attached: = en, <anexado: = pt, <archivo adjunto: = es). Pass as whisper code: en, pt, es, etc.
whatsapp-name-conversation/
.gitignore # ignores exported/
.processed # metadata (transcription preference, language)
chat-2024-01-01_2024-06-15.md
chat-2024-06-16_2024-12-31.md
exported/ # gitignored — full export contents
_chat.txt
00000001-PHOTO-*.jpg
00000002-AUDIO-*.opus
00000002-AUDIO-*.txt # transcriptions live alongside audio
**John:**
[08/04/2026, 14:30:00] Hello there
[08/04/2026, 14:31:00] [Photo](exported/00000001-PHOTO-abc.jpg)
[08/04/2026, 14:32:00] 🎤 "Transcribed voice message text here"
> Audio: [exported/00000003-AUDIO-abc.opus](exported/00000003-AUDIO-abc.opus)
[08/04/2026, 14:33:00] [Document: contract.pdf](exported/00000005-DOC-contract.pdf)
[08/04/2026, 14:34:00] [Voice message](exported/00000004-AUDIO-def.opus)
**Jane:**
[08/04/2026, 14:35:00] Got it, thanks!
Messages are grouped by sender — the name only appears when the sender changes. Non-transcribed voice messages appear as [Voice message](exported/AUDIO-xxx.opus).
ls *.md in the output folder shows date ranges at a glance.txt file in exported/${CLAUDE_SKILL_DIR}/scripts/check-deps.sh
Run this before processing to verify all tools are available.
Transcription uses whisper.cpp locally. Fully offline — no API keys needed.
First-time setup (requires ffmpeg and cmake — install with brew install ffmpeg cmake):
Check if whisper.cpp is already installed (~/.whisper.cpp/build/bin/whisper-cli or locally in the skill directory). If not found, ask the user whether to install shared (~/.whisper.cpp, reusable across projects) or local (inside the skill directory). Then run:
${CLAUDE_SKILL_DIR}/scripts/setup.sh --shared # installs to ~/.whisper.cpp
${CLAUDE_SKILL_DIR}/scripts/setup.sh --local # installs to skill directory
This clones whisper.cpp, builds it, and downloads the large-v3 model (~3GB). Guide the user through installation if prerequisites are missing.
Performance: ~2-10 seconds per voice message on Apple Silicon.
The parser handles multiple languages for attachment markers:
<attached: filename><anexado: filename><archivo adjunto: filename>PHOTO-) sent right after messages mentioning amounts