Claude Code Setup for AI Implementation Experts

# CLAUDE.md — AI Implementation Expert & Second Brain Architect
> Version: 2026-03 | Difficulty: advanced | This is a real setup powering a live product

## What Your AI Can Now Do
- Ship production features autonomously with `/yalla` — 4-agent planning team with adversarial red-team review before a single line of code is written
- Auto-track every outreach touchpoint (WhatsApp, email, LinkedIn) to CRM via hooks — zero manual logging
- Generate personalized Second Brain repositories for clients, fully automated from questionnaire to GitHub delivery
- Get a morning intelligence briefing combining sales pipeline, inbox, and task priorities with one `/overview` command
- Draft, evaluate, and publish LinkedIn posts with a single `/content-generation` command — voice consistency scored against 1,000+ real posts
- Run parallel agent teams: 1 lead + 3 specialists working simultaneously on different layers of the same feature

---

## Quick Start
1. `/begin [task]` before ANY code change — hooks enforce session branch discipline automatically
2. `/yalla` for complex features — plans with a 4-agent team, then builds and reviews before shipping
3. `/overview` every morning — real data from CRM, tasks, and calendar combined
4. `/respond-second-brain` for multi-channel outreach (WhatsApp + Gmail + LinkedIn in one command)
5. `/grow` weekly — brain health score, patterns learned, time savings metrics

---

## About This Setup
I run a multi-product business (SaaS, coaching, open-source MCP) while doing enterprise innovation advisory. My AI handles product development, CRM automation, content production, and client delivery simultaneously. The goal: think once, let the system run.

**What I'm optimizing for:**
- Zero context-switching between tools — everything surfaces in Claude
- Every decision leaves a trace (memory, task, commit)
- Automation that degrades gracefully — manual fallback for every automated step

---

## Pre-Built Commands

| Command | What It Does |
|---------|-------------|
| `/yalla` | Full autonomous pipeline: plan → build → test → review → ship |
| `/begin [desc]` | Start tracked session with branch + GitHub issue auto-created |
| `/end` | Close session with PR, merge, branch cleanup |
| `/overview` | Morning briefing: pipeline status + tasks + comms digest |
| `/respond-second-brain` | Multi-channel outreach: WhatsApp + Gmail + LinkedIn |
| `/content-generation` | Generate, evaluate, and queue LinkedIn/newsletter content |
| `/publish-blog` | Full pipeline: write → evaluate → fix AI slop → SEO → publish |
| `/add-prospect` | Add to CRM with auto-LinkedIn enrichment |
| `/grow` | Brain health score and learning metrics |

---

## How Claude Connects to Your Tools

| Tool | Why You Need It |
|------|----------------|
| Supabase MCP | Primary database: CRM, blog posts, product data |
| Resend MCP | Transactional email: purchase confirmations, drip sequences |
| Stripe MCP | Payment history, customer lookup, subscription management |
| Gmail (gog CLI) | Email outreach, inbox triage, reply drafting |
| LinkedIn (AnySite MCP) | Prospect research, profile enrichment, post engagement |
| Google Analytics MCP | Traffic analysis, content performance |
| Claude in Chrome MCP | LinkedIn automation, browser-based flows |
| Memory MCP | Persistent knowledge graph — context survives between sessions |
| GitHub MCP / gh CLI | Issues, PRs, file access without browser |

---

## Rules

- **MUST:** Run `/begin` before any code change — hooks block edits on main branch, no exceptions
- **MUST:** Use `mcp__supabase__execute_sql` for all DB queries
- **MUST:** Include `.js` extensions in all API imports (production build requirement)
- **MUST:** Add new routes BEFORE the catch-all `*` route
- **MUST:** Confirm before destructive actions (force push, hard reset, table drops)
- **SHOULD:** Use `Write` (not sequential `Edit` calls) for files over 500 lines — 80% fewer token failures
- **SHOULD:** Run `npm run check:types && npm run test:unit` before every push

---

## Project Structure
```
src/pages/          # React pages
api/                # Vercel serverless functions
lib/                # Shared utilities
.claude/hooks/      # Lifecycle automation (branch-guard, CRM sync)
.claude/docs/       # Architecture, patterns, gotchas
memory/             # Semantic patterns, personal profile, episodic sessions
product/            # Vision docs, specs, context engineering
```

---

## Automation (Hooks)

| Trigger | What Happens |
|---------|-------------|
| Any edit on `main` | `branch-guard.sh` blocks it — must be on session branch |
| Session start | Auto-switches to `main`, prevents branch reuse |
| WhatsApp send | Auto-syncs to CRM with timestamp and message summary |
| `git push` | Runs `check:types + test:unit` automatically |

---

## Gotchas
- **Missing `.js` in API imports** → Build fails in production
- **Route after catch-all `*`** → 404 forever, route never matches
- **Code after `res.json()`** → Vercel terminates function — use `waitUntil()` for background work
- **Large files (500+ lines) + many small Edits** → Use `Write` instead — saves 80% of tokens
- **Agent Teams on main branch** → Run `/begin` first — hooks apply to teammates too

---

## Memory & Learning
Claude builds context over time using the Memory MCP:
- Architectural decisions and why they were made
- Client patterns and buying signal recognition
- Content voice calibration (1,000+ LinkedIn comments analyzed)
- Recurring bugs and their root causes
- Patterns learned across sessions stored in `memory/semantic/patterns/`

Save anything: just say "remember that [fact]" in conversation.
Review: `/grow` shows health score + patterns learned this week.