Claude Dev Stack

One command to give Claude Code persistent memory, git conventions, and a complete dev workflow.

npx claude-dev-stack

Claude Code is powerful — but it forgets everything between sessions. Claude Dev Stack fixes that.

v1.0.0-alpha.1 (Pre-release)

Alpha channel — install via npm install -g claude-dev-stack@alpha. The stable @latest tag (v0.12.x) is unchanged.

v1.0 introduces the CDS-Core architecture: a pnpm monorepo with Claude Agent SDK integration, SQLite session memory, and automatic session capture.

What's new in alpha:

• Auto session capture — sessions are automatically saved to a per-project SQLite database. No more /end required.
• SQLite memory layer — FTS5 full-text search over session observations, entities, and relations.
• MCP adapter — sessions.search, docs.search, planning.status tools available directly inside Claude Code.
• `/cds-quick` — one-shot agent dispatch with cost reporting.
• Backfill migration — claude-dev-stack migrate sessions ports historical markdown sessions into SQLite.
• CC 2.x permission hardening — claude-dev-stack doctor --gsd-permissions configures the GSD executor allowlist.

Links:

• Migration guide — breaking changes + rollback instructions
• Changelog — full list of additions, changes, deprecations

Requirements: Node 20+ (Node 18 no longer supported). See the migration guide for details.

The Problem

If you use Claude Code seriously, you've hit these walls:

• Total amnesia. Every session starts from zero. Claude doesn't remember what you worked on yesterday, what decisions you made, or what's left to do.
• No git discipline. Claude writes commit messages like fix stuff or update code. No scopes, no conventional commits, no branch conventions — unless you manually instruct it every time.
• Manual setup per project. You copy-paste the same CLAUDE.md, install the same plugins, configure the same MCP servers. Every. Single. Project.
• Scattered documentation. Session notes, architecture decisions, project context — all lost in chat history that Claude can't access next time.

The Solution

Claude Dev Stack sets up an Obsidian-compatible vault that stores everything Claude needs to resume work across sessions — context, decisions, session logs, git conventions — and auto-loads it every time.

Before vs After

Quick Start

1. Install

npx claude-dev-stack

The interactive wizard walks you through:

• Vault creation (where Claude stores memory)
• Project discovery and mapping
• Component selection (skills, hooks, plugins, MCP servers)
• Git conventions setup per project
• Stack-specific context.md generation

2. Fill in context.md

This is the single most important step. Claude reads this file at the start of every session.

cd ~/Projects/my-project && claude
> "help me fill in context.md for this project"

Describe your project's purpose, tech stack, architecture, current state, and known issues. The more specific you are, the better Claude performs across sessions.

3. Start working

cd ~/Projects/my-project && claude
> "let's continue"

Claude automatically loads your project context, shows what was done last session, and proposes next steps. When you're done:

> "done for today"

Claude creates a session log with what was done, decisions made, and TODO for next time.

Use Cases

Session Continuity

You close your laptop Friday evening. Monday morning:

> "let's continue"

Project: my-saas-app
Last session (2026-04-11): Implemented user auth flow, added JWT refresh.
TODO: Wire up email verification endpoint, add rate limiting.

Suggested next step: Email verification — the auth flow is complete
but unverified users can still access protected routes.

No context lost. No "remind me what we were doing."

Git Conventions

Claude Dev Stack detects your monorepo structure and generates scopes automatically:

claude-dev-stack scopes init --quick
# Detected: pnpm-workspace (apps/web, apps/api, packages/ui)
# Generated .claude/git-scopes.json with scopes: web, api, ui

Now when you say "commit this as a fix":

fix(api): resolve database connection timeout on cold start

Not fix stuff. Supports 7 monorepo types: pnpm, npm/yarn workspaces, Nx, Turborepo, Lerna, Cargo workspaces, Go multi-module, and Python uv.

NotebookLM Research

Sync your vault to Google NotebookLM for docs-grounded research:

claude-dev-stack notebooklm sync
# Syncing 3 projects to per-project notebooks...
# cds__my-saas-app: 12 sources (3 new, 9 unchanged)
# cds__mobile-app: 5 sources (1 updated)
# cds__shared-lib: 2 sources (all current)

Each project gets its own cds__{slug} notebook. Hash-based delta sync — only changed files are re-uploaded. Sync runs automatically on every session end — your notebooks stay current without manual intervention.

Query your notebooks directly from Claude:

> "ask notebooklm what we decided about the auth flow"

Answer: In session 2026-04-11, you decided to use JWT with refresh tokens...
Citations: [sessions/2026-04-11-auth-flow.md, decisions/0003-jwt-auth.md]

Or from the CLI:

claude-dev-stack notebooklm ask "what's our API rate limiting strategy?"
claude-dev-stack notebooklm generate report    # AI-generated summary of your project
claude-dev-stack notebooklm generate mind-map  # Visual mind map of project knowledge

Upgrading from v0.8? Migrate existing sources safely:

claude-dev-stack notebooklm migrate           # dry-run — shows what would move
claude-dev-stack notebooklm migrate --execute  # two-phase-commit: upload+verify, then delete from shared

Notion Auto-Import

Pull Notion docs into your vault without leaving the terminal:

claude-dev-stack notion add https://www.notion.so/workspace/API-Spec-abc123
claude-dev-stack notion import

Imported files land in vault/projects/{slug}/docs/notion/ with provenance stamps. If you edit the imported file locally and Notion gets updated, re-import writes a .notion-update.md sibling instead of overwriting your changes.

In a live Claude session, you can also trigger imports via the notion-importer skill using MCP — no CLI needed.

Multi-Project Workflow

Working across multiple projects in a session:

> "switch to mobile-app"
# Saves current context, loads mobile-app context

> "what did we decide about the auth flow?"
# Reads from mobile-app/decisions/ — finds ADR-003

> "switch back to my-saas-app"
# Restores previous context seamlessly

What's Included

CLI Reference

Setup

claude-dev-stack                           # Full interactive setup wizard

Projects

claude-dev-stack projects                  # List projects with status
claude-dev-stack projects add              # Add a project to vault
claude-dev-stack projects remove           # Remove project from vault

Git Conventions

claude-dev-stack scopes init               # Initialize git-scopes.json (--quick or --full mode)
claude-dev-stack scopes list               # Show current scopes and config
claude-dev-stack scopes add <name>         # Add a scope
claude-dev-stack scopes remove <name>      # Remove a scope
claude-dev-stack scopes refresh            # Re-detect scopes from project structure

Notion Import

claude-dev-stack notion list               # Show configured Notion pages
claude-dev-stack notion add <url>          # Add Notion page to import config
claude-dev-stack notion import             # Import all configured pages (or --page <id>)

NotebookLM

claude-dev-stack notebooklm sync           # Sync vault to per-project NotebookLM notebooks
claude-dev-stack notebooklm status         # Show sync status, file counts, stale files
claude-dev-stack notebooklm ask "question" # Query your notebook — returns answer + citations
claude-dev-stack notebooklm generate report     # Generate AI summary of project knowledge
claude-dev-stack notebooklm generate mind-map   # Generate visual mind map
claude-dev-stack notebooklm migrate        # Dry-run migration from shared to per-project notebooks
claude-dev-stack notebooklm migrate --execute  # Execute migration (two-phase-commit)

Documents

claude-dev-stack docs                      # List documents per project
claude-dev-stack docs add                  # Add from files, Notion export, or paste

Skills

claude-dev-stack skills                    # List installed skills
claude-dev-stack skills install            # Install from catalog or Git URL
claude-dev-stack skills remove             # Remove installed skills

Plugins

claude-dev-stack plugins                   # List installed plugins
claude-dev-stack plugins install           # Install via preset or browse all 98+
claude-dev-stack plugins presets           # Install curated set (fullstack, frontend, backend, etc.)
claude-dev-stack plugins marketplaces      # Add third-party marketplaces (19 available)

MCP Servers

claude-dev-stack mcp                       # List configured MCP servers
claude-dev-stack mcp install               # Install from catalog (18 servers)
claude-dev-stack mcp remove                # Remove MCP servers

Templates

claude-dev-stack new                       # Generate context.md from stack template

14 stacks: Next.js, Nuxt, SvelteKit, Astro, React+Vite, FastAPI, Express, Laravel, Spring Boot, Rails, Django, Flutter, Go, Blank.

Import & Export

claude-dev-stack import                    # Import AI configs from a project directory

Supports: CLAUDE.md, .cursorrules, .windsurfrules, copilot-instructions.md, .clinerules, .aider.conf.yml.

claude-dev-stack export                    # Export vault as .tar.gz
claude-dev-stack sync init                 # Initialize vault as git repo
claude-dev-stack sync push                 # Commit and push vault
claude-dev-stack sync pull                 # Pull latest from remote

Maintenance

claude-dev-stack doctor                    # Health check for all components
claude-dev-stack update                    # Update skills, hooks, GSD, CLI
claude-dev-stack stats                     # Analytics dashboard
claude-dev-stack help                      # Show all commands
claude-dev-stack version                   # Show version

Vault Structure

~/vault/
├── meta/
│   ├── project-registry.md          # All mapped projects
│   └── session-protocol.md          # Session conventions
├── shared/
│   ├── patterns.md                  # Reusable patterns across projects
│   └── infra.md                     # Infrastructure notes
├── research/                        # Deep Research outputs
├── projects/
│   ├── my-project/
│   │   ├── context.md               # ★ Project context (the most important file)
│   │   ├── sessions/                # Auto-generated session logs
│   │   ├── decisions/               # Architecture Decision Records
│   │   └── docs/                    # Documentation
│   │       └── notion/              # Notion auto-imports (with provenance stamps)
│   └── _template/
└── CLAUDE.md.template

How Skills Work

Skills are markdown files that tell Claude how to behave in specific situations. They auto-trigger based on what you say — no slash commands needed.

Two hooks run at the Claude Code level for reliability (skills can be ignored by Claude, but hooks always fire):

For Teams

Every team member runs:

npx claude-dev-stack

To share the vault:

claude-dev-stack sync init     # Initialize vault as git repo
# Add remote, push to shared repo
# Each team member clones/pulls the vault

Session logs, ADRs, and context stay in sync. Each developer's Claude sessions contribute to a shared knowledge base.

Dev Containers

Works inside VS Code dev containers and GitHub Codespaces. See examples/devcontainer/ for a complete configuration.

{
  "image": "mcr.microsoft.com/devcontainers/javascript-node:22",
  "mounts": [
    "source=${localEnv:HOME}/.ssh,target=/home/node/.ssh,type=bind,readonly",
    "source=${localEnv:HOME}/.claude,target=/home/node/.claude,type=bind"
  ],
  "postCreateCommand": "npm install -g @anthropic-ai/claude-code && npx --yes claude-dev-stack"
}

With vault git sync, your context follows you between host and container automatically.

Updating

npx claude-dev-stack update

Updates built-in skills, hooks, git-based skills (Obsidian, Deep Research), GSD, and Claude Code CLI.

Requirements

• Node.js 18+ — runtime
• Git — for vault sync and skill installation
• Claude Code CLI — for plugins and MCP server configuration
• Python 3.10+ (optional) — for Deep Research and NotebookLM sync (notebooklm-py)

The wizard detects missing tools and shows install commands:

✘ node — not found
  ℹ Install: brew install node

Credits

Built on top of:

• GSD (Get Shit Done) by TÂCHES — spec-driven development
• Obsidian Skills by kepano — vault file format support
• Deep Research Skills by Weizhena — structured research
• notebooklm-py by teng-lin — NotebookLM CLI wrapper

Support

If this saves you time, consider buying me a coffee:

License

MIT