Claude Code Skills para Fluxos de Reunião
OxyFlax/AI
Transcrição, documentação e migração de conversas para fluxos de reunião
Ver no GitHubColeção de skills e agents para Claude Code focada em automação de reuniões: transcrição de áudio, normalização de volumes, geração automática de documentação e sincronização de projetos. Inclui também ferramentas de migração de históricos de Copilot e Antigravity, com processamento de áudio/vídeo 100% local (Apple Silicon).
Instalação
npx convez-ai OxyFlax/AI1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Transcrever e documentar automaticamente reuniões em markdown
- Migrar históricos de conversas do GitHub Copilot ou Google Antigravity
- Normalizar níveis de áudio em gravações multipessoas
- Sincronizar atualizações de reuniões com documentação de projetos
Veredito

Skill maduro e bem documentado para times que usam Claude Code intensivamente. O processamento on-device é um diferencial forte para privacidade. Requer setup com ffmpeg e dependências Python, mas cobre um gap real em automação de reuniões.
README
A tradução para o português deste README ainda está na fila.
Claude Code Skills
A collection of Claude Code skills and agents for meeting workflows (recording normalization, transcription, documentation generation, keeping project docs in sync) and tool migrations (moving conversation history and project context from Copilot, Antigravity, and similar tools into Claude Code).
All audio and video processing runs locally on-device (Apple Silicon). No audio or video is sent to the cloud.
Skills
| Skill | What it does |
|---|---|
antigravity-migration | Extract, document, and continue Google Antigravity conversations and brain artifacts |
audio-normalize | Balance voice levels in multi-speaker recordings |
copilot-migration | Extract, document, and continue GitHub Copilot Chat conversations |
meeting-transcribe | Transcribe audio to text, locally with faster-whisper |
meeting-sync | Process a transcript and update all project docs |
meeting-to-documentation | Generate markdown docs + screenshots from a video recording |
spark-prompt | Generate a GitHub Spark app prompt from an idea |
translate-docs | Translate markdown files into another language |
Prerequisites
Different skills have different dependencies. Install only what you need:
# Required by: audio-normalize, meeting-transcribe, meeting-to-documentation
brew install ffmpeg
# Required by: meeting-transcribe
pip install faster-whisper # or: see skills/meeting-transcribe/references/setup.md
# Optional: local on-device transcription via whisper.cpp (audio-normalize --transcribe)
brew install whisper-cpp
Installation
Copy the skill folders you want into your Claude Code project or global skills directory, then restart Claude Code.
skills/
├── antigravity-migration/
├── audio-normalize/
├── copilot-migration/
├── meeting-transcribe/
├── meeting-sync/
├── meeting-to-documentation/
├── spark-prompt/
└── translate-docs/
Skills Reference
antigravity-migration
Extracts your Google Antigravity agent conversations and brain artifacts (task lists, implementation plans, walkthroughs) from disk (~/.gemini/), decodes them into readable Markdown transcripts, copies brain artifacts verbatim, exports project permission grants and MCP config, and helps you continue your most recent conversation in Claude Code.
Trigger phrases: "extract my Antigravity conversations", "migrate from Antigravity", "recover my agent history", "continue my last Antigravity conversation"
/antigravity-migration --list
/antigravity-migration --out antigravity-history --filter my-project
Produces:
antigravity-history/transcripts/— faithful Markdown transcripts (encrypted conversations noted)antigravity-history/artifacts/<conversation>/— brain artifacts copied verbatimantigravity-history/sessions/— per-conversation summaries (written by Claude)antigravity-history/README.md— chronological index and thematic groupingantigravity-history/CONTINUE_HERE.md— handoff for resuming your most recent conversationantigravity-history/antigravity-settings.json— exported permission grants and MCP config
Also migrates durable project context into CLAUDE.md, maps permission grants into .claude/settings.json, and translates mcp_config.json to .mcp.json.
Everything runs locally and read-only — no data leaves your machine. Requires python3 (standard library only).
audio-normalize
Balances voices in a multi-speaker recording where some participants are far from the microphone. Uses a three-stage ffmpeg chain (high-pass filter → dynamic compression → EBU R128 loudness normalization to −16 LUFS). Output is a _normalized.m4a in the same folder.
Trigger phrases: "normalize this recording", "balance the voices in this audio", "prepare this m4a for transcription"
/audio-normalize ~/Downloads/meeting.m4a
/audio-normalize ~/Downloads/meeting.m4a --transcribe
Optionally runs local whisper.cpp transcription after normalization (--transcribe flag).
copilot-migration
Extracts your GitHub Copilot Chat and agent-mode conversation history from disk, documents them as readable Markdown with summaries and an index, and helps you continue your most recent conversation — bringing your full Copilot context into Claude Code without losing anything.
Trigger phrases: "extract my Copilot conversations", "migrate from GitHub Copilot", "recover my chat history", "continue my last Copilot conversation"
/copilot-migration --list
/copilot-migration --out copilot-history
Produces:
copilot-history/transcripts/— faithful, verbatim Markdown transcriptscopilot-history/sessions/— per-conversation summaries (you write these)copilot-history/README.md— chronological index and thematic groupingcopilot-history/CONTINUE_HERE.md— handoff for resuming your most recent conversation
Everything runs locally and read-only — no audio or chat data leaves your machine.
meeting-transcribe
Transcribes an audio file to a timestamped markdown transcript using faster-whisper (large-v3) with built-in VAD (Silero). Runs locally on Apple Silicon — no API key required. Optimized for French and English.
Trigger phrases: "transcribe this recording", "generate a transcript", "turn this voice memo into text"
/meeting-transcribe ~/Downloads/meeting.m4a
Produces:
transcripts/<basename>.transcript.md— human-readable with timestampstranscripts/<basename>.transcript.json— raw segments for downstream tooling
The .md file is the expected input for meeting-sync.
meeting-sync
Processes a meeting transcript (text or file path) and updates all project documentation in one pass:
| File | Action |
|---|---|
docs/meeting-notes/meeting-YYYY-MM-DD.md | Created — structured English recap |
docs/project-overview.md | Updated if new strategic or architectural facts emerged |
docs/backlog.md | Updated — new items added, decisions recorded, open questions updated |
Trigger phrases: "sync meeting notes", "process this transcript", "update the backlog from this meeting"
/meeting-sync path/to/transcript.md
/meeting-sync "paste the raw transcript text here"
The skill reads existing docs before writing, so it only modifies sections touched by the transcript.
meeting-to-documentation
Turns a video recording (.mp4) plus a transcript (.vtt, .srt, or .txt) into a polished markdown document with embedded screenshots. Uses ffmpeg scene-change detection to extract candidate frames, correlates them with the narration, curates the best set (5–15 screenshots), and writes documentation that reads as if authored directly — no mention of the recording or session.
Trigger phrases: "generate docs from this recording", "meeting to documentation", "turn this walkthrough into a README"
/meeting-to-documentation VIDEO=/path/demo.mp4 TRANSCRIPT=/path/demo.vtt OUTPUT_DIR=/path/output
Produces:
output/documentation.mdoutput/screenshots/(curated, renamed frames)
All intermediate artifacts are cleaned up automatically.
spark-prompt
Generates a compelling, feature-rich prompt for GitHub Spark from a plain-language app idea. The output is a single, self-contained prompt ready to paste into Spark — no implementation instructions, no framework mentions.
Trigger phrases: "generate a Spark prompt", "write a Spark app prompt", "create a no-code app prompt"
/spark-prompt A personal finance tracker for freelancers
translate-docs
Translates .md documentation files into a target language (default: French) and places the output in a language-named folder mirroring the source path. Supports incremental updates — when the translation already exists, only changed sections are re-translated.
Trigger phrases: "translate this doc", "create a French version", "update the translation"
/translate-docs docs/project-overview.md
/translate-docs docs/backlog.md spanish
Output convention: docs/foo.md → docs/french/foo.md.
Typical Meeting Workflow
Voice Memo (.m4a)
│
▼
/audio-normalize ← balance voices if needed
│
▼
/meeting-transcribe ← produce timestamped .md transcript
│
▼
/meeting-sync ← update all project docs from the transcript
For recorded video walkthroughs (demos, feature reviews):
Recording (.mp4) + Transcript (.vtt)
│
▼
/meeting-to-documentation ← markdown doc + curated screenshots
Agents
The agents/ directory contains project-specific agents built on top of these skills. They are meant as examples — adapt them to your own project structure and documentation conventions.
License
MIT