i am still not sure, using Claude, Codex, Deepseek, GLM, all work great..😁

Thank you for this video "Build Your Own AI Software Factory with Claude Code" and for sharing the full source code! I'm not a developer, but the factory was simple enough that I could understand it and put it to work on my own project, a small FastAPI service that turns YouTube videos into English transcripts. It has now run 12 jobs on it, from spec to build to review to merge. The approver escalating Docker changes to me was exactly the safety net I needed. One improvement I made, in case it's useful to others: I upgraded the code reviewer using ideas from Matt Pocock's code-review skill. It now reviews in two separate passes: 1. Spec: every requirement in spec.md is marked met, partly met or missing, with evidence from the diff. 2. Standards: the repo's own rules, plus a baseline of 12 Fowler code smells (duplicated code, speculative generality, feature envy and so on). Smells only go under Notes, so they never block a job or waste review rounds. It still starts with VERDICT: PASS or CHANGES, so nothing else in the factory had to change. The reviews are now much easier to trust, because you can see how each requirement was checked instead of just reading "matches the spec". I also ran a one-off audit of the whole codebase and turned the findings into backlog items, and /factory next worked through them one by one. Thanks again. This was the most practical Claude Code video I've watched. for sharing you repo.. attached my modified reviewer-code.md

loved the Software Factory, Implemented the same, work as designed.