From cm
Syncs high-value dev knowledge (skills, decisions, lessons) into Google NotebookLM for cloud recall, podcasts, and flashcards. Pairs with Dockit/deep-search.
How this skill is triggered — by the user, by Claude, or both
Slash command
/cm:cm-notebooklmThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Cloud brain / soul cho AI — lưu trữ kiến thức **giá trị nhất**, đồng bộ qua
Cloud brain / soul cho AI — lưu trữ kiến thức giá trị nhất, đồng bộ qua
nhiều máy tính. brain.md = compiled source of truth, upload 1 lần thay vì 50.
Skills + Lessons + Experiences
↓ compile
~/.codymaster/brain.md (local file = backup)
↓ upload (1 API call)
NotebookLM (cloud = cross-machine sync)
All automation via scripts/brain-sync.sh — AI chỉ cần gọi script.
CodyMaster uses a Dual-Brain approach to scale memory across multiple projects without context collapse:
codymaster): Global brain storing Meta-Learnings, proven Architecture Decisions, and universal patterns. Evolved via the "Rule of 3" (knowledge graduation)../docs.| Level | Goal | Query Command |
|---|---|---|
| Master | Tìm pattern chung | nlm notebook query codymaster "..." |
| Project | Tìm context domain | nlm notebook query $(cat .cm/notebook_id) "..." |
🤖 AI Rule: Tự đưa quyết định khi hỏi: Hỏi
codymastercho các giải pháp mang tính hệ thống chung. Hỏi qua.cm/notebook_idcho các business logic đặc thù dự án đang làm.
# 1. Install CLI
uv tool install notebooklm-mcp-cli
# 2. Setup Master Brain
bash ~/.gemini/antigravity/skills/cm-notebooklm/scripts/brain-sync.sh init
Nếu dự án đủ lớn và nhiều doc:
# Tạo Project Brain riêng cho thư mục hiện tại
bash ~/.gemini/antigravity/skills/cm-notebooklm/scripts/brain-sync.sh init-project
SCRIPT=~/.gemini/antigravity/skills/cm-notebooklm/scripts/brain-sync.sh
# Add a lesson learned
bash $SCRIPT lesson "Tên bài học"
# → Edit ~/.codymaster/lessons.md → fill in details
# Add coding experience
bash $SCRIPT experience "Tên pattern"
# → Edit ~/.codymaster/experiences.md → fill in details
# Sync to Master Brain (Thêm rule of 3)
bash $SCRIPT sync
# Sync to Project Brain (Up tài liệu local docs/)
bash $SCRIPT sync-project
# Check status
bash $SCRIPT status
# Query
nlm notebook query codymaster "your question"
AI detect và hỏi user (không tự động):
| Trigger | Prompt |
|---|---|
| Skill mới tạo | "Sync skill mới vào brain?" → bash $SCRIPT sync |
| Bug fixed / post-mortem | "Lưu bài học?" → bash $SCRIPT lesson "..." |
| Architecture changed | "Update brain?" → bash $SCRIPT sync |
| User nói "sync/update" | → bash $SCRIPT sync |
uv tool install notebooklm-mcp-cli
nlm login
nlm notebook list # CodyMaster Brain already there
nlm alias set codymaster <id>
# Done! Query ngay: nlm notebook query codymaster "..."
nlm audio create codymaster --format deep_dive --confirm # Podcast
nlm report create codymaster --format "Study Guide" --confirm
nlm flashcards create codymaster --difficulty medium --confirm
nlm studio status codymaster
✅ HIGH-VALUE (auto-compiled into brain.md):
├── Skill Index (names + descriptions — NOT full SKILL.md)
├── Lessons Learned (~/.codymaster/lessons.md)
├── Coding Experiences (~/.codymaster/experiences.md)
└── AGENTS.md (project identity)
❌ NOT INDEXED (use qmd/cm-deep-search instead):
├── Full SKILL.md files (too many, hard to maintain)
├── Source code, tests, configs
└── Duplicated content
Session → variables → temporary
Working → CONTINUITY.md → ~500 words/turn
Local → qmd → BM25+vector, offline, stable
Cloud → NotebookLM → AI brain, cross-machine, podcast
| Skill | Role |
|---|---|
cm-deep-search | Local search complement (code) |
cm-dockit | Generate docs → select high-value → feed to brain |
cm-continuity | Session memory, brain = long-term |
skill-creator-ultra | TRIGGER: new skill → prompt sync |
cm-debugging | TRIGGER: bug fixed → prompt lesson |
nlm CLI = third-party (jacob-bd). May break. Fallback: cm-deep-search.nlm chat start — use nlm notebook query only.npx claudepluginhub tody-agent/codymaster --plugin cmCaptures insights as markdown files, searches prior learnings, and promotes patterns to CLAUDE.md using tiered backends (local, qmd, agent-fs) for knowledge across projects.
Captures patterns, decisions, gotchas, procedures, and feature knowledge from conversations into durable skills for on-demand reuse. Invoked via /learn or 'please remember'.
Captures cross-project learnable patterns (decisions, errors, insights) into a persistent semantic graph via Neural Memory MCP. Auto-recalls context at session start and captures learnings after feature work, debugging, or code review.