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 :)