PRODUCT • Claude Code optimization
Your Claude Code is operating at 40% capacity.
Most devs treat Claude like a chatbot. The ones building autonomous empires treat it like an engineer with rules. This is the exact CONSTITUTION + hooks + skills setup that runs my 300-job casino-affiliate empire 24/7.
Why this exists: I built this for myself over 3 months. Friends saw my output and asked. I'm packaging it because if it pays my debt, someone else's debt deserves a faster path too.
What you get
- CONSTITUTION.md template — 8 inviolable P1 rules + 6 strong-default P2 rules. Pre-resolves the conflicts between MARCUS.md, SOUL.md, CLAUDE.md. Copy-paste ready.
- 26 production hooks — money filter, no-menu-options, no-punt, no-theater, no-hedge, smoke-test-required, ground-truth-probe, more. Each one blocks a specific failure mode.
- 15 production skills — /premortem, /verify-work, /goal, /ultrareview equivalents. Each one is a complete prompt + matcher.
- Memory system — Letta/MemGPT-style file-based memory that survives context resets. Auto-organized by feedback / project / reference / user types.
- Self-improvement loop — Claude builds its own gates as it learns. Memory rule violated 3x = automatic code-gate generation.
- Empire-grade orchestration — supervisor → worker patterns. Council queue. Coverage audits. Self-healers. All productionalized for macOS launchd.
BEFORE
- Claude forgets the rule you told it 5 min ago
- Asks "should I continue?" instead of executing
- Hedges with "I think this should work"
- Punts to "next session" / "for later"
- Claims success without verifying
- Recommends paid services to indebted users
AFTER
- Rules become hooks that throw — can't skim past them
- Auto-executes goal-aligned next steps
- Hedge phrases blocked at the Stop hook
- "Next session" language fires no-punt detector
- Smoke-test required before any "done" claim
- Money filter throws on paid recommendations
Built from 3 months of real failures
Every hook in this pack was generated AFTER my Claude failed the same way 3+ times. The CONSTITUTION exists because Claude kept making the same architectural mistake. Each rule has a date + reason + cost-of-failure logged. This isn't theory — it's scar tissue.
What it's NOT
- Not a course or a 2-hour video lecture
- Not Anthropic-affiliated
- Not a fork of someone else's "awesome-claude-code" repo
- Not a one-time setup — includes the meta-pattern for evolving your own rules over time
FAQ
Does this work with Claude API too, not just Claude Code CLI?
Yes. Hooks are bash scripts that work anywhere you call Claude. Skills work in Claude.ai web too. The CONSTITUTION pattern works for any agent framework.
Will Anthropic's next update break this?
Hooks API has been stable since Q1 2026 and is publicly documented. Skills are first-class. Memory system is file-based (yours, not Anthropic's).
Refund?
If your Claude doesn't ship at least one feature within 24h of installing this, full refund. I'm betting it will.