Code Factory: a clearer path from AI-generated code to reviewed work

I build Code Factory. I’m working on the gap between getting an AI to write code and having enough evidence to accept its work.

The idea is one local-first factory with three connected roles:

SpecLine: clarify the request, write the spec and plan, and approve manageable tasks.

ForgeLine: move through implementation, review, tests, and smoke checks with recorded receipts.

Code Factory: scan and audit the code, collect evidence, and show findings and next actions.

Imagine asking your coding agent to add CSV export. Before it starts, you decide whose records can be exported and what an empty export should do. Then you work through the tasks and relevant checks, keeping a record of what was checked against which inputs.

I want that process to help newer coders understand the next step and help senior developers review an agent’s work without reconstructing the whole conversation. Supporting smaller models through clearer tasks and checks is also a design goal; I haven’t established parity with frontier models.

You can begin with a scan of an existing Python repository and add the broader workflow as needed. The scanner is advisory, with known false positives. Its receipts support review; they don’t certify software or replace tests and human judgment.

Code and getting started: https://github.com/zrk222/code-factory

What would make this useful in your own workflow: clearer task boundaries, easier-to-review evidence, or more explicit next actions?