English · Русский · Español · Português (BR) · Türkçe · 中文 · 日本語
7 products in 4 months · solo · pre-revenue
I started writing code in January 2026 with Cursor and Claude. Four months later: 3 live sites, 4 SaaS codebases ready to deploy, 1 active YouTube channel. This repo is the stack and the playbook I run from.
- What this is
- The Stack
- The 7 products in 4 months
- Quick Start
- What's Inside
- The Operator Playbook
- Cookbook
- Solo-founder skills (originals)
- Slash commands
- Scaffolds
- Profiles
- Why this exists
- How this compares
- Long-form docs
- Acknowledgements
- Status
- License
A curated toolkit and a playbook for solo founders running several AI products at once.
The stack is what I install and update. The playbook is how I use it across the week — which components to reach for in which order, what to read first, where the seams are.
Most parts of the stack are other people's work, credited where used. What's added here: the install path, the workflows that compose the parts together, and four case studies of products built on top.
It's aimed at people running 2+ products at the same time, founders without a CS background, and anyone who wants Claude Code to do real work instead of being a chat companion.
Core (always install):
| Layer | Component | Author | What it does for me |
|---|---|---|---|
| Orchestration | Claude Code | Anthropic | The runtime |
| Second Brain | Obsidian | Obsidian | ~/Brain vault as project + identity context |
| Knowledge graph | graphify | local | Folder of files → navigable knowledge graph with community detection |
| UI generation | Frontend-Design | Anthropic | Distinctive, production-grade UI |
Opt-in (install when the use case fits):
| Layer | Component | Author | When to add |
|---|---|---|---|
| Skills + Agents | Everything Claude Code | @affaan-m | When you want a broad skill + agent catalog (182 skills, 48 agents) |
| SEO + Ads | Toprank | nowork-studio | When you do SEO audits or run Google/Meta Ads |
Every skill and agent in this stack credits its original author. If a piece comes from somewhere else, the link goes there.
See stack/ for component-by-component setup notes.
What this stack actually shipped between January and May 2026.
| # | Product | Status | Stack |
|---|---|---|---|
| 1 | Niche Booking Trio — 3 niche booking sites | Live (3 domains) | Next.js · Supabase · GA4 · Sentry |
| 2 | P2P Marketplace — P2P classic-car rental | Code complete | Next.js · Stripe Connect · Prisma |
| 3 | WhatsApp B2B SaaS — WhatsApp SaaS for dealers | Code complete | FastAPI · Docker · WhatsApp Cloud API |
| 4 | AI Legal Tool — AI traffic-fine appeals | Code complete | Next.js · Prisma · Claude API |
| 5 | YouTube production pipeline | Live (active) | Python · yt-dlp · Whisper · Claude |
| 6 | Jarvis Workspace — personal AI assistant | Live (daily use) | Claude Code · Obsidian · graphify |
| 7 | Internal ops automation | Live | hooks + skills + cron |
See case-studies/ for the how.
Sets up the stack on a fresh machine. macOS and Linux supported; Windows via WSL.
Pick one install path. Don't run both back-to-back on a fresh
~/.claude/— they target the same files and the second run will create duplicate sidecars.
git clone https://github.com/mccarthy606/claude-operator-stack.git
cd claude-operator-stack
less install.sh # audit it first
./install.sh --dry-run # preview every change
./install.sh # applyThe installer will:
- Verify
claudeCLI is installed (and abort with instructions if missing) - Print the marketplace + plugin commands you'll run inside Claude Code
- Copy sanitized
settings.jsonandmcp-servers.jsontemplates to~/.claude/as sidecar files — your existing config is never silently overwritten - Print the next-step checklist for adding your own API keys
Nothing is committed to your ~/.claude/ without explicit confirmation. The installer supports --dry-run and --yes flags.
Available after the package is published in Phase 9. The npm registry returns 404 until the public-launch flip ships
npm publish. Use the bash path above until then.
npx claude-operator-stack init --dry-run # preview
npx claude-operator-stack init # apply
npx claude-operator-stack verify # audit your existing setup
npx claude-operator-stack list-stack # show the wired componentsSame outcome as install.sh, different ergonomics. The wizard prompts you through marketplace selection, copies sanitized configs as sidecar files (*.from-operator-stack), and prints the manual /plugin commands you'll run inside Claude Code.
The repo is a four-layer toolkit: a stack/ of curated components, a set of workflows/ that compose them, content artefacts (cookbook/, case-studies/, scaffolds/, profiles/, skills/), and supporting machinery (configs/, commands/, docs/, tests/, credits/). Top level:
claude-operator-stack/
├── stack/ ← 4 core + 2 opt-in component breakdowns
├── workflows/ ← 5 operator playbooks
├── case-studies/ ← 4 anonymised shipped products
├── cookbook/ ← 12 copy-pasteable recipes
├── skills/ ← 6 original SKILL.md packages
├── commands/ ← 6 slash-commands wrapping the skills
├── scaffolds/ ← web-saas + whatsapp-saas
├── profiles/ ← 4 archetype install paths
├── packages/cli/ ← npm CLI sibling to install.sh
├── configs/ ← sanitized settings/hooks/rules
├── docs/ ← long-form guides + extractions
├── tests/ ← E2E integration suite
└── credits/ ← attribution to every original author
Five workflows that run my week.
From idea to live URL in one session. See workflows/ship-a-product-in-a-day.md.
Keeping seven projects in flight without losing context between them. See workflows/parallel-projects.md.
Every project also has a note in ~/Brain; Claude Code reads it on session start. See workflows/obsidian-as-context.md.
 |
 |
| Left: the project note Claude reads at session start. | Right: Claude loading that context, then asking where to pick up. |
YouTube, Instagram, and drive2 across three brands with most of the production steps automated. See workflows/content-pipeline.md.
Customer support, billing, scheduling, and infra handled from one person's calendar. See workflows/solo-ops.md.
Twelve copy-pasteable recipes pulled out of real shipped products. Each one ≤200 lines: problem, solution, code, pitfalls.
A few examples:
- Stripe Connect onboarding for a P2P marketplace
- WhatsApp Cloud API webhook in FastAPI
- Cloudflare Tunnel for local-dev webhooks
- Sentry across Next.js + FastAPI in one project
Full index in cookbook/README.md.
Six original SKILL.md packages targeting solo-founder use-cases ECC's 182-skill catalog doesn't cover. They ship alongside ECC, not instead — both directories coexist under ~/.claude/skills/.
| Skill | Use case |
|---|---|
solo-billing-monitor |
Weekly cost rollup across cloud + AI APIs |
multi-project-context-bridge |
Bridge cross-project decisions via graphify queries with anonymisation |
obsidian-sync-helper |
Verify Brain notes vs git state |
case-study-anonymiser |
Apply the redaction playbook to a draft case study |
weekly-monday-review |
Monday review → 2-of-N focus pick |
ship-day-planner |
One-line idea → 8 ship-day blocks |
Cookbook recipes are how-to docs the operator reads; skills are invocable prompts Claude executes when triggered. See skills/README.md.
Six slash-commands that wrap the six original skills as one-keystroke verbs in Claude Code's slash-picker. Each command parses operator-friendly args, applies sensible defaults, and delegates to its wrapped skill for the protocol. The skill is the source of truth; the command is the shorthand.
| Command | Use case | Wrapped skill |
|---|---|---|
/solo-monday-review |
Monday 30-min ritual → 2-of-N focus pick | weekly-monday-review |
/anonymise-case-study |
Apply the redaction playbook to a draft | case-study-anonymiser |
/ship-day |
One-line idea → 8 ship-day blocks | ship-day-planner |
/cost-rollup |
Weekly cross-cloud + AI cost rollup | solo-billing-monitor |
/bridge-context |
Bridge cross-project decisions via graphify, anonymised | multi-project-context-bridge |
/sync-brain |
Verify Brain note vs git state | obsidian-sync-helper |
These coexist with Everything Claude Code's 67 commands under ~/.claude/commands/. Names checked at plan-time and execute-time against ECC's catalog — zero collisions. See commands/README.md for the full index.
Two runnable starting points. Clone the directory, fill in .env, ship.
scaffolds/web-saas/— Next.js 15 + Supabase + Sentry + GA4 with a real lead form, real/api/leadroute, and a pre-configuredCLAUDE.mdtuned for the stack.scaffolds/whatsapp-saas/— FastAPI + Docker + Meta Cloud API + Anthropic SDK with HMAC verification, a Claude classifier, and a happy-path pytest.
Both ship with placeholder CLAUDE.md blocks (visual direction, project name, etc.) marked deliberately rather than fake-looking defaults.
Pick the install path that matches how you describe yourself. Each profile picks ~6 cookbook recipes, 2-4 hooks, a scaffold, and a workflow read order — and tells you what to skip.
- Indie hacker — solo dev shipping 2-4 small bets in parallel
- Non-technical founder — Claude is your engineer, you do the framing
- Freelancer / small agency — N client repos with shared baseline + per-client overrides
- Content creator + operator — content is half your income, software the other half
Index in profiles/README.md.
Most AI-tooling material is written for engineers. This is written for operators.
The bet: a non-engineer with a tight project list, a curated stack, and a workflow that compounds can ship more than a small team, given the right setup. I am not trying to argue AI replaces engineers; I am documenting what one operator can do with the right tools loaded.
Most components here are other people's work. What's mine is the glue: the install path, the workflows, and the case studies that pull seven separate projects into one stack.
Solo Stack is the operator-first wrapper around Claude Code; Everything Claude Code (ECC) is the upstream skills + agents catalogue this repo depends on; starter templates are framework-first scaffolds. Solo Stack and ECC are designed to coexist — many readers will install both.
Depth artefacts that don't fit the README's first-impression job: the full comparison table, the annotated tree, a narrative changelog, and the rationale for the single-harness scope.
See docs/README.md for the index.
Built with:
- @affaan-m — Everything Claude Code (skills and agents)
- nowork-studio — Toprank (SEO, Google Ads, Meta Ads)
- Anthropic — Claude Code, Frontend-Design, the API
- Obsidian team — the second-brain runtime
- Every skill author credited individually in
origin:frontmatter and in credits/README.md
If your work is in here and not credited, open an issue and I'll fix it the same day.
Young repo. v1.0 added the cookbook (12 recipes), scaffolds (web + WhatsApp), profiles (4 archetypes), 6 hooks, the ECC skill index, and screenshots. The post-v1.0 wave ([Unreleased]) adds 6 original SKILL.md packages, an npm CLI sibling to install.sh, an animated SVG hero, and a stack reframe to 4 core + 2 opt-in components alongside the OMEGA → graphify knowledge-graph migration. Earlier v0.2 added the hero banner, Mermaid diagrams, 7-language nav, and full RU/ES translations (RU/ES coverage of the v1.0 + post-v1.0 content is tracked as an open issue). Case studies get filled in as products ship. CHANGELOG tracks the rest.
Issues, PRs, and forks welcome. The stack is designed to be customized: copy what fits, drop what doesn't.
Open good first issues if you want to contribute:
- Translate README to PT-BR · TR · 中文 · 日本語 (stubs in place, replace with full translations)
- Add native Windows install script (
install.ps1) - Sync RU + ES translations with v1.0 content (cookbook, scaffolds, profiles, screenshots)
- Add Mermaid to content-pipeline.md (~30 min)
See all open issues and CONTRIBUTING.md.
MIT. See LICENSE.
The components this stack depends on each carry their own licenses — see each component's repo and credits/README.md.