Most "AI agent packs" are speculative prompts written in an afternoon. This one runs a real business. The Claude Code Production Pack is the complete, sanitized .claude configuration behind wowhow.cloud — a live e-commerce operation with 4,000+ digital products, 150+ free browser tools, and roughly 490 published articles, operated end-to-end by Claude Code by a solo founder. You are not buying ideas about how AI agents might work. You are buying the operating system of a store that exists, with the scar tissue included.
What you get
26 specialist agent definitions. A full virtual company: an architect that plans with dependency graphs, a builder that implements to spec, a reviewer, and — the one buyers copy first — the verification-agent: an adversarial QA agent that tries to BREAK every change and returns a PASS/FAIL/PARTIAL verdict before anything ships. Around them sits a 12-persona growth department (SEO auditor, CRO surgeon, analytics oracle, content architect, competitive intel operative, performance engineer, monetization strategist, support triage, growth coordinator), plus a security scanner, a deployment agent, a multi-agent orchestrator, and a financial computing engine. Each file defines persona, tool permissions, and an output contract — not vibes.
14 original workflow skills. Repeatable procedures encoded so any session executes them identically: tool-builder (ship a production browser tool end-to-end), blog-writer (three content modes with SEO rules and an anti-AI-slop enforcement list), seo-fix-loop (Search Console data to content fix to verification), prod-verify (never claim "deployed" without proof from production), persistent-planner, product-creator, marketplace-lister (lists products on Gumroad through the official API with ban-avoidance drip pacing), medium-publisher, pinterest-dominator, wrapup (end-of-session memory that compounds week over week), and more.
The rules are the real product
Six rule files distilled from real production incidents. The trust-boundary rule blocks any commit touching payment, auth, or webhook code until a written security checklist is acknowledged — it exists because a payment bug shipped without it. The deploy-invariants file holds 31 hard rules, each one written the day a deploy broke something. The social rate-limit rule exists because a platform suspended an account after 190 automated posts — the API allowed it; the moderators didn't. The anti-AI-writing file bans the exact phrases and structures that make content read like a bot wrote it. These are lessons that normally cost weeks of downtime and lost revenue, encoded so you skip them.
Study the templates
The pack includes the production CLAUDE.md and AGENTS.md (sanitized) as reference templates. Study how a real operation encodes model tiering (which model handles which task class, and when to escalate), failure laws, a decision engine, and a definition of done that requires actual command output — not "looks good" — before anything is called finished. Adapt the structure to your project in an afternoon.
Who this is for
Developers and solo founders using Claude Code who want their AI to operate like a disciplined team instead of an eager intern. If you have ever had an agent claim something was done when it wasn't, deploy something that broke, or write content that screamed AI — the specific countermeasures for all three are in this pack.
Details
108 files. Works with Claude Code CLI and desktop, any current model tier. Install by copying into your project's .claude/ directory or user-wide — the included GETTING-STARTED guide is a 10-minute read with a recommended gradual adoption path. Single-user license, unlimited projects, commercial use included. Instant download, pay once, keep forever.
Using gstack? They compose. gstack (Garry Tan's open-source setup) covers the engineering workflow — code review, QA, release. This pack covers what it deliberately doesn't: the business layer — growth, marketing automation, revenue recovery, marketplace distribution, and commerce-hardened rules for payments, SEO, and deploys. Run both.