Right after the launch of Kiro, I started using it actively. I was really drawn to the idea of specification-driven development—a clear requirements pipeline, detailed instructions for the AI agent, and a shared structure understandable by both developers and other team members.
But in practice, Kiro turned out to be too immature:
— It frequently froze and ran slowly,
— Generated low-quality output and odd requirements like “As a developer, I want logging in the project,”
— Kept trying to force its own chaotic vision of testing onto my workflow.
Eventually, I got tired of fighting these limitations and switched back to my familiar Cursor. But there, I had to rebuild the whole process from scratch: teaching the agent what user stories are, what system design means, the context of the current project, and much more. It became clear that I needed a prompt library—a set of foundational instructions to set the rules of the game from the start.
So I gathered everything I could:
— Reverse-engineered prompts from Kiro,
— Recalled best practices from books and team documentation I’d written over the years,
— And merged it all into a single, cohesive guide.
The result is the Specification-Driven Development Guide: a ready-to-use context for LLMs that you can plug into any agent or chat interface right away.
This is, of course, just the first version. It still needs significant refinement—especially the English translation, which I’m not entirely confident about. But I hope it’s already useful to someone.
https://methodology.uwuwu.net/ - feedback is very welcome!
Top comments (0)