Project : ForgeWorks

An AI coding agent will happily write a demo. It will not, on its own, write something you can ship. ForgeWorks is the difference: one command turns an empty folder into a project with planning, a real test pyramid, a security review, and a second reviewer built in.

It is not a starter app. It installs the rules and the gates. The agent has to plan before it writes, write the failing test before the code, and pass an independent review before anything merges. A solo builder gets the discipline of a senior team.

The rules live in one file that every major agent reads, so they travel. Two models review instead of one, because two perspectives catch what one misses. Ceremony is sized to the change: a typo goes straight through, a new behaviour does not.

This site was built with it.

What it is made of

A Claude Code multi-agent engineering system that turns an idea into tested, security-gated software across four language stacks. Every feature goes through the same loop: design, failing tests, implementation, an independent review pass, and a merge that will not happen while a gate is red.

This site was built with it in 2026. That is the honest proof: the factory and its output are both here to read.

Why it exists

I was programming alone and could see what was missing. Not more code: a team.

I write something, and it needs review. It needs tests. The technical debt has to be paid. It has to be deployed. And when I use the thing I built, the feedback needs somewhere to land, to be ranked, and to get done.

That is a software factory, and no single person is one.

ForgeWorks takes the problem off me, proposes the solution, builds what the solution needs, and then measures how much of the problem actually went away.