Story

Hi, a little context for this simple project:

I only started this "agentic development" thing this year thanks to the Google AI Pro sale, therefore my most used agent is Antigravity - Gemini 3, which is smart but a loose cannon. So my attempt to "tame" this model turned into this. (Btw I have claude code and cursor)

It could be useful for people who:

- are used to conventional coding and want to try out agentic development

- don't vibe-code or prefer staying in the loop

- using less polished harness or less rigorous models

It's a skill defining a methodology to command LLMs. You reference the workflow files to make LLM plan, execute, review, red-team, etc. — each output a typed markdown document with a defined template and lifecycle.

Documents chain together (Plan → Execute → Audit), are just files in your repo, and double as persistent memory across sessions. Basically a lightweight project management protocol between you and the agent.

I am posting this simply because it's proven useful for me; I wouldn't say this is good or smart way to approach agentic development :)