AI continuous development - My Toilet-to-Prod Pipeline

The Itch I've always liked playing with scripts, CI, and automating things. But this time, with LLMs, I think I outdid myself. I built a workflow that starts as a Telegram chat, which helps me shape the specs for a feature that launches as a GitHub issue. From there, I use a Docker image for planning changes, implementing code, running reviews, and merging code when quality gates and the CI build pass. And then CI releases a new version, but this is the boring part. I had an unhealthy amount of fun connecting all those things together, and I still am tweaking it to my liking. I know there are out-of-the-box agent systems that just pick your prompt and plan work on a kanban board, use worktrees for parallel development, and generate specs using one of the frameworks (what a bold name for a markdown file). I tried some and didn’t like it. It was burning tokens to provide a generic experience, with the same approach to a TypeScript backend app, React frontend, native mobile, and GoLang. S