← All work

AI Developer Infrastructure — Independent build

claudepwn

One command turns a product spec into a fully configured, multi-agent Claude Code environment — ready to plan, build and harden itself.

Role

Independent build

Ships as

npx claudepwn

Generates

8 agents · 10 commands · 7 rules · 9 skills

Stacks

Python · TypeScript · Rust · Go

claudepwn terminal Multi-Agent · Model Routing · 8-Wave Hardening Claude­pwn

From a PRD to a running multi-agent build in a single command — planning on Opus, implementing on Sonnet, closing on Haiku.

What it does

Given a product-requirements document and a target stack, claudepwn scaffolds a complete .claude/ environment, then has Claude read the PRD headless, decompose it into epics and sprints, write architecture-decision records, seed a documentation vault, and kick off the first sprint. For existing codebases it runs an audit first, opens a hardening sprint, then plans forward.

What I built

  • Eight specialised agents — architect, implementer, tester, reviewer, security, docs, vault-keeper, dispatcher — coordinating asynchronously over a shared status board.
  • Ten commands, seven coding rules, four lifecycle hooks and nine skills, generated as ~80 canonical files.
  • Per-language stack overlays wiring the right lint / test / format / audit tools for Python, TypeScript, Rust and Go.
  • An eight-wave hardening gate: pre-check → tests → security scan → code review → dependency analysis → docs sync → domain safety → PR review.
  • A sprint runner that routes each phase to the right model tier for cost, and an optional Obsidian vault for ADRs and design knowledge.

How it's built

A thin Node CLI wraps a Bash orchestrator that scaffolds non-destructively, detects the stack from PRD keywords, and launches Claude headless with bootstrap prompts. Agents don't lock — they coordinate through append-only board and dispatch files, so work runs in parallel without deadlocks. A pre-commit hook regex-scans for secrets before anything reaches git, and an optional GitNexus knowledge-graph integration maps module boundaries for impact analysis.

Why it matters

It's the operating system for how I actually ship — the same multi-agent, hardened, model-routed cadence that produced most of the work on this site, packaged so anyone can run it.

Node.jsBashClaude / Anthropic SDKMulti-agentModel routingGitNexus

Core feature

From claudepwn's actual codebase

The real CLI in action: `npx claudepwn new` scaffolds a complete .claude/ environment — 85 files across eight agents, ten commands, seven rules, four hooks, nine skills, a vault and plugins — in one command, then hands off to Claude.

npx claudepwn new — real scaffold run

claudepwn — zsh — 100×40 npx claudepwn
~/projects/limen-socimi $ npx claudepwn new --prd spec.md --features all Project name from PRD: limen-socimi Stack from PRD: typescript ═══════════════════════════════════════════════════ claudepwn new ═══════════════════════════════════════════════════ Project: limen-socimi Stack: typescript PRD: spec.md Features: vault checklists sprints gitnexus plugins monorepo ═══════════════════════════════════════════════════ ▸ Copying base .claude/ structure… ▸ Applying typescript overlay… ▸ Copying PRD… ▸ Filling placeholders… ▸ Creating vault/… ▸ Creating checklists/… ▸ Creating scripts/… ▸ Setting up GitNexus… ▸ Creating monorepo scaffold… ▸ Installing plugins… security-guidance@anthropics-claude-code manual code-review@anthropics-claude-code manual frontend-design@anthropics-claude-code manual claude-mem@thedotmack manual ═══════════════════════════════════════════════════ Scaffold complete! 85 files. ═══════════════════════════════════════════════════ 8 agents · 10 commands · 7 rules · 4 hooks · 9 skills — .claude/ ready.