Not a prompt collection. Not a cheat sheet. A battle-tested methodology for solo founders who ship.
77 specialist skills — each one loaded only when its expertise is needed. Drop in the React expert when building UI. Activate the RLS Guardian when writing migrations. Zero token bloat.
A structured skill framework for Claude Code with SQLite-backed persistent memory, semantic vector search, deterministic hooks, always-on rules, and slash commands.
MemStack gives Claude Code persistent memory across sessions, semantic recall via vector search, automated safety checks on every commit and push, portfolio governance to prevent over-engineering, work verification, and context compression to make your sessions last longer.
- TTS voice notifications — Cross-platform text-to-speech (Windows, macOS, Linux). Pre-prompt alerts fire BEFORE approval prompts so you know to return to the terminal.
- Diary webhook — Session logs POST to n8n endpoint after markdown backup. Fire-and-forget so webhook failure never blocks saves.
- Pro skills catalog —
pro-catalog.mdrule provides organic upsell: mentions relevant Pro skills once per session when a task matches.
- Governor (#19) — Portfolio governance with 3-tier system (Prototype/MVP/Production). Prevents over-engineering by enforcing tier-appropriate complexity.
- Description trap audit — All 17 skill descriptions rewritten to say WHEN to invoke, never HOW the skill works. Prevents Claude from shortcutting full protocols.
- Anti-rationalization tables — Echo, Diary, and Verify now include tables of known Claude excuses with rebuttals, improving protocol compliance.
- Silent context compilation — Work skill (Lv.5) now silently gathers STATE.md, CLAUDE.md, recent diary, and git state before any plan operation.
- Patterns adopted from Intellegix Code Agent Toolkit and Superpowers plugin research.
- Humanize (#16) — Remove AI writing patterns from text output. Curated pattern table + voice guidelines.
- State (#17) — Living STATE.md document tracking current task, blockers, and next steps. Auto-reads at session start.
- Verify (#18) — Pre-commit verification reports. Checks build, tests, and requirements before committing.
- Seal upgrade — Commit format now supports conventional commits (
feat(scope): description) alongside[ProjectName]format. - Diary upgrade (Lv.5) — Structured Session Handoff section for seamless pickup between sessions.
- Echo upgrade (Lv.5) — Semantic vector search via LanceDB with SQLite fallback (added in v3.0.1).
MemStack is a lightweight alternative to heavyweight frameworks like GSD — same capabilities, pure markdown, zero dependencies.
- Claude Code — Install guide: https://docs.anthropic.com/en/docs/claude-code
- Python 3.10+ — Download: https://www.python.org/downloads/
- Git — Download: https://git-scm.com/downloads
- LanceDB + sentence-transformers (optional, for semantic recall) —
pip install lancedb sentence-transformers
MemStack™ follows a Progressive Disclosure pattern:
- Skills load lean (~80 lines of context)
- The MCP Skill Loader surgically fetches only what your exact task needs
- Full expert depth activates on demand — not dumped into every session
- Result: zero token bloat, full specialist precision
One command. Works across every project immediately.
git clone https://github.com/cwinvestments/memstack .claude/skillsClone the full repo, then link it into each project:
git clone https://github.com/cwinvestments/memstack.git
cd memstackcp config.json config.local.jsonOpen config.local.json and edit the projects section with your actual paths:
Windows: "path": "C:\\Projects\\my-app"
Mac/Linux: "path": "/home/user/projects/my-app"
Everything else works with defaults.
python db/memstack-db.py initCopy the .claude/ folder into any project you want MemStack to manage:
Windows:
xcopy /E /I /Y C:\Projects\memstack\.claude C:\Projects\yourproject\.claudeMac/Linux:
cp -r /path/to/memstack/.claude /path/to/yourproject/.claudeThat's it. Start Claude Code in your project directory and begin working — hooks fire automatically, rules load every session, and skills activate on matching triggers. No activation line needed.
Without
.claude/installed: If you're in a project that doesn't have MemStack's.claude/folder, you can still use skills manually by adding this to your prompt:Read C:\Projects\memstack\MEMSTACK.md and follow the MemStack skill framework.
LanceDB + sentence-transformers enables semantic vector search for the Echo skill — finding relevant past sessions by meaning, not just keywords.
pip install lancedb sentence-transformersIndex your existing sessions:
python skills/echo/index-sessions.pyEcho will automatically use semantic search when LanceDB is available, and fall back to SQLite keyword search when it's not.
For higher-quality embeddings, optionally set an OpenAI API key:
Windows: setx OPENAI_API_KEY sk-...
Mac/Linux: echo 'export OPENAI_API_KEY=sk-...' >> ~/.bashrc && source ~/.bashrc
Headroom compresses tool outputs by ~34%, making sessions last longer.
pip install headroom-aiMemStack auto-detects and auto-starts it. To make permanent:
Windows: setx ANTHROPIC_BASE_URL http://127.0.0.1:8787
Mac/Linux: echo 'export ANTHROPIC_BASE_URL=http://127.0.0.1:8787' >> ~/.bashrc && source ~/.bashrc
MemStack includes one-click launcher scripts that start Headroom, verify it's healthy, and open VS Code.
| Platform | File | How to run |
|---|---|---|
| Windows | start-memstack.bat |
Double-click, or right-click → Send to → Desktop (create shortcut) |
| Mac/Linux | start-memstack.sh |
chmod +x start-memstack.sh then ./start-memstack.sh |
What the launcher does:
- Checks if Headroom is already running on port 8787 (skips start if so)
- Starts the Headroom proxy if needed and waits for initialization
- Runs a health check to confirm the proxy is responding
- Opens VS Code to your projects directory
If Headroom is not installed, the launcher reports the failure and continues — it's optional.
MemStack's rules, hooks, and commands live in .claude/. Link this folder into any project using a junction (Windows) or symlink (macOS/Linux) so updates propagate instantly.
First-time setup:
git clone https://github.com/cwinvestments/memstack.git C:\Projects\memstackLink a project (Windows):
start-memstack.bat link C:\Projects\YourProjectOr manually:
mklink /J C:\Projects\YourProject\.claude C:\Projects\memstack\.claudeLink a project (macOS/Linux):
ln -s /path/to/memstack/.claude /path/to/YourProject/.claudeHow it works: The junction/symlink makes your project's .claude/ point to MemStack's .claude/. Any updates to rules, hooks, or commands in MemStack instantly apply to all linked projects — no copying or re-syncing needed.
Removing a link:
| Platform | Command | Notes |
|---|---|---|
| Windows | rmdir C:\Projects\YourProject\.claude |
Removes the junction only, not the source |
| macOS/Linux | rm C:\Projects\YourProject\.claude |
Removes the symlink only, not the source |
MemStack v3.2.3 uses three layers with increasing reliability:
MemStack v3.2.3
├── Hooks (deterministic) — Shell scripts, CC lifecycle events
│ ├── pre-push.sh — Build check, secrets scan, commit format (standard + conventional)
│ ├── post-commit.sh — Debug artifacts, format validation
│ ├── session-start.sh — Headroom auto-start + CLAUDE.md indexer + API monitor
│ └── session-end.sh — Report "completed" to monitoring API
├── Rules (always-loaded) — Markdown files, loaded every session
│ ├── memstack.md — Global conventions, commit format, deprecated skill guard
│ ├── echo.md — Memory recall protocol (vector + SQLite)
│ ├── diary.md — Session logging protocol (with handoff)
│ ├── work.md — Task planning protocol
│ └── headroom.md — Compression proxy awareness
├── Commands (slash) — Quick-access utilities
│ ├── memstack-search.md — /memstack-search <query>
│ └── memstack-headroom.md — /memstack-headroom (proxy stats)
└── Skills (context-aware) — Markdown protocols, keyword triggers
├── Echo, Diary, Work — SQLite + LanceDB vector memory (Lv.5)
├── Governor — Portfolio governance (Lv.1) ← NEW in v3.2
├── State, Verify, Humanize — Workflow quality (Lv.1)
├── Project, Grimoire — Session lifecycle (Lv.2-3)
├── Scan, Quill — Business tools (Lv.2)
└── Forge, Shard, Sight — Dev tools (Lv.2)
Hooks always fire (deterministic, zero tokens). Rules always load (persistent awareness). Skills fire on keyword/condition match.
| Command | What It Does |
|---|---|
/memstack-search <query> |
Search SQLite memory for past sessions, insights, and project context |
/memstack-headroom |
Check Headroom proxy status and token savings |
| Skill | Emoji | Level | What It Does |
|---|---|---|---|
| Echo | 🔊 | Lv.5 | Semantic recall via LanceDB vectors + SQLite fallback |
| Diary | 📓 | Lv.5 | Documents sessions to SQLite + structured handoff for seamless pickup |
| Work | 📋 | Lv.5 | Plan execution with SQLite-backed task tracking + silent context compilation |
| Governor | 🏛️ | Lv.1 | Portfolio governance — tier/phase constraints prevent over-engineering |
| Project | 💾 | Lv.3 | Saves/restores project state via SQLite context |
| Humanize | ✍️ | Lv.1 | Removes AI writing patterns — makes text sound human |
| State | 📍 | Lv.1 | Living STATE.md — tracks current task, blockers, next steps |
| Verify | ✅ | Lv.1 | Pre-commit verification — checks build, tests, requirements |
| Grimoire | 📖 | Lv.2 | Manages CLAUDE.md files across projects |
| Familiar | 👻 | Lv.2 | Splits tasks across multiple CC sessions |
| Scan | 🔍 | Lv.2 | Analyzes project scope and suggests pricing |
| Quill | ✒️ | Lv.2 | Generates professional client quotations |
| Forge | 🔨 | Lv.2 | Creates new MemStack skills |
| Shard | 💎 | Lv.2 | Refactors large files into smaller modules |
| Sight | 👁️ | Lv.2 | Generates Mermaid architecture diagrams |
| Compress | ⚙️ | Lv.1 | Manages Headroom proxy — status, stats, troubleshooting |
Deprecated skills (Seal, Deploy, Monitor) have been replaced by deterministic hooks that always fire — no LLM required.
All memory is stored in SQLite (db/memstack.db) with WAL mode:
python db/memstack-db.py search "authentication" # Search everything
python db/memstack-db.py get-sessions my-app # Recent sessions
python db/memstack-db.py get-insights my-app # Decisions and patterns
python db/memstack-db.py stats # Database overview| Table | Purpose |
|---|---|
sessions |
Session diary entries (written by Diary, read by Echo) |
insights |
Extracted decisions and patterns |
project_context |
Current state of each project (auto-indexed from CLAUDE.md) |
plans |
Task lists with per-task status |
config.local.json sections:
| Section | Required | Purpose |
|---|---|---|
projects |
Yes | Your project paths and CLAUDE.md locations |
headroom |
No | Context compression settings (port, auto-start) |
cc_monitor |
No | External dashboard API URL and key |
session_limits |
No | Max lines for session log exports (default: 500) |
defaults |
No | Commit format, auto-diary, auto-monitor toggles |
git pull
python db/memstack-db.py init # Update database schema
python db/migrate.py # Import existing markdown files (safe to re-run)
python db/memstack-db.py stats # Verify migrationUse the Forge skill: say "forge a new skill for [description]" in any CC session. Forge generates the file with YAML frontmatter and registers it in the master index.
"Python was not found" — Install Python 3.10+ and make sure "Add Python to PATH" is checked during install. Restart your terminal after installing.
CC sessions not using Headroom — Run headroom proxy in a separate terminal first, then launch CC. Or install permanently with setx ANTHROPIC_BASE_URL http://127.0.0.1:8787 (Windows).
Hook errors on push — The pre-push hook blocks pushes that contain secrets or fail build checks. Fix the issue it reports, then push again.
"Database locked" errors — Close any other process accessing db/memstack.db, or delete the .db-wal and .db-shm files and retry.
The free repo includes 77 skills (17 core + 60 categorized). MemStack™ Pro adds 3 Pro-exclusive skills — bringing the total to 80 across 10 categories — plus an MCP Skill Loader, a semantic search MCP server where Claude Code calls find_skill("your task") and loads only the relevant skill on demand.
Core (17) — Memory, planning, workflow, and utilities
| Skill | Description |
|---|---|
| Diary | Session logging with git integration and SQLite storage |
| Echo | Past session recall via semantic vector search + SQLite |
| Work | Task planning and todo list management |
| State | Load and update project context at session start |
| Project | Save project state and handoff context |
| Verify | Verification before committing completed work |
| Governor | Project maturity assessment and complexity budgeting |
| Grimoire | Update project context files after significant changes |
| Compress | Headroom proxy status and token savings |
| Humanize | Make AI-generated text sound natural |
| Forge | Create new MemStack skills |
| Familiar | Dispatch work across parallel CC sessions |
| Scan | Codebase complexity analysis and project estimation |
| Quill | Generate client-facing price quotations |
| Shard | Split and manage files over 1000 lines |
| Sight | Visual diagrams and architecture overviews |
| KDP Format | Convert manuscripts to KDP-ready .docx |
Security (7) — Audits, policies, vulnerability scanning
| Skill | Description |
|---|---|
| RLS Checker | Supabase Row Level Security policy verification |
| RLS Guardian | RLS enforcement on new/altered database tables |
| API Audit | API endpoint protection verification |
| OWASP Top 10 | Comprehensive web security review against OWASP Top 10 |
| Secrets Scanner | Exposed secrets detection in source code |
| Dependency Audit | Vulnerability scanning and abandoned package detection |
| CSP Headers | HTTP security headers (CSP, HSTS, X-Frame-Options) |
Deployment (6) — CI/CD, containers, hosting, DNS
| Skill | Description |
|---|---|
| Railway Deploy | Application deployment to Railway with env vars and domains |
| Netlify Deploy | Static site and serverless function deployment to Netlify |
| Docker Setup | Container optimization with Dockerfile and docker-compose |
| CI/CD Pipeline | Automated build, test, and deployment pipelines (GitHub Actions) |
| Domain SSL | DNS records, SSL certificates, and custom domain configuration |
| Hetzner Setup | VPS provisioning, hardening, and deployment |
Development (7) — Architecture, testing, performance
| Skill | Description |
|---|---|
| Database Architect | Supabase/Postgres table structures, relationships, and RLS |
| API Designer | RESTful API route design with request/response schemas |
| Code Reviewer | Structured code quality, security, and performance reviews |
| Test Writer | Unit, integration, and component tests with mocking |
| Migration Planner | Safe schema evolution with zero-downtime strategies |
| Performance Audit | Frontend and backend performance diagnosis and optimization |
| Refactor Planner | Systematic code improvement and tech debt reduction |
Business (6) — Proposals, contracts, invoicing
| Skill | Description |
|---|---|
| Proposal Writer | Project proposals for client and freelance engagements |
| Scope of Work | Project boundaries, deliverables, and acceptance criteria |
| Contract Template | Professional service contracts with legal clauses |
| Client Onboarding | Structured onboarding process for new clients |
| Invoice Generator | Professional invoices with line items and payment instructions |
| Financial Model | Financial projections with scenario modeling and unit economics |
Content (8) — Blog, email, video, social
| Skill | Description |
|---|---|
| Blog Post | Long-form written content for blogs and publications |
| Email Sequence | Multi-email automated campaigns with nurture sequences |
| Landing Page Copy | Persuasive short-form copy for product landing pages |
| Newsletter | Newsletter editions with subject lines, content, and growth strategy |
| Product Description | Conversion-optimized product descriptions for e-commerce |
| TikTok Script | Scripts with hooks and visual cues for 15–60s vertical videos |
| Twitter Thread | Multi-tweet narratives with hooks, data points, and CTAs |
| YouTube Script | Scripted content for YouTube with hooks, chapters, and CTAs |
SEO & GEO (6) — Search optimization, structured data
| Skill | Description |
|---|---|
| Site Audit | Website SEO health evaluation |
| Keyword Research | Target keywords with search volume and difficulty |
| Meta Tag Optimizer | HTML meta tag optimization for search visibility |
| Schema Markup | Schema.org structured data (JSON-LD) for rich results |
| AI Search Visibility | Content optimization for AI-powered search engines |
| Local SEO | Local search optimization (Google Business Profile, NAP) |
Marketing (8) — Funnels, ads, launches, pricing
| Skill | Description |
|---|---|
| Sales Funnel | Complete customer journey from stranger to repeat buyer |
| Facebook Ad | Social media ad copy with targeting for Meta platforms |
| Google Ad | Keyword groups, headlines, and Quality Score optimization |
| Launch Plan | Day-by-day launch timeline with pre/post checklists |
| Competitor Analysis | Pricing, feature, and messaging comparisons against competitors |
| Pricing Strategy | Pricing tiers, psychology application, and A/B testing |
| Lead Magnet | Lead capture assets with landing page copy and nurture sequences |
| Webinar Script | Timestamped presentation scripts with slide notes |
Product (6) — PRDs, specs, roadmaps, backlog
| Skill | Description |
|---|---|
| PRD Writer | Engineering-ready PRD with problem statement and personas |
| Feature Spec | Detailed feature specifications with user flows and edge cases |
| User Story Generator | Prioritized stories with Given/When/Then criteria |
| MVP Scoper | Smallest viable build that validates product hypothesis |
| Roadmap Builder | Strategic planning with themes, milestones, and resources |
| Feedback Analyzer | Support ticket and review categorization and prioritization |
Automation (5) — Workflows, webhooks, scheduling
| Skill | Description |
|---|---|
| n8n Workflow Builder | Design automated workflows with triggers and data transformations |
| Webhook Designer | Secure webhook receivers with validation and idempotency |
| Cron Scheduler | Recurring background jobs with monitoring and failure handling |
| API Integration | Build reliable connections between systems via their APIs |
| Content Pipeline | Automate creation, formatting, and publishing across platforms |
| Skill | Description | Unlock |
|---|---|---|
| 🔒 Consolidate | Weekly cross-project summaries and pattern detection | Pro |
| 🔒 Context DB | SQLite-backed facts database — query project knowledge instead of reading full CLAUDE.md | Pro |
| 🔒 API Docs | Fetch current API documentation via Context Hub before writing API code | Pro |
MIT — see LICENSE for details.
Copyright (c) 2026 CW Affiliate Investments LLC