GEMINI.md's, Skills, \doc and what goes where - Best Pratice

I have been using AntiGravity for a couple of months for pure Vibe Development. Recently started using MCP servers for my Postgres, Redis, and Git work and then last week, Antigravity persuaded me to also start using skills.

I’ve spent quite a lot of effort making sure that my GEMINI.md (both global and workspace), is optimized the way I work. Updating them as I implement new features and bring in new components to my tech stack. I have semi-automated this with my code reviews/feature_complete, suggesting improvements for the GEMINI.md files.

But with the addition of skills, I saw a lot of overlap with what should go into Skills, GEMINI.md’s and also my regular “for humans to read” documentation.

So now we (Claude and me) have created yet another maintenance prompt “config-layer-audit.md”, that is optimizing both GEMINI.md, Skills as well as cross reference it with regular documentation in my \doc folders.

To me it makes sense, but I am not sure if its best practice to work in Antigravity like this.

  • So, will it actually make my AI’s perform better when coding for me?
  • What is best practice for this? What are you doing?

Note that i am 100% Vibe Developing and OK with that not being cheap, so really do not care that Skills save tokens not filling the context window with everything on every prompt. I care more about quality, getting it right. Speed is of course also nice, but with Antigraivity’s agent manager to have AIs work in parallel, really not a problem.

Below you can see my setup. And as you can see, I am starting to have quite a lot of “AI Config/AI DevOps/Maintenance” prompts. (I even have META prompts optimizing my review prompts, to fit the code base as it evolves, but that’s another story). Anyway, I find myself spending more and more time prompting my AIs to do other stuff than writing code to implement new and enhanced features.

Hello,

Thank you for fantastic to see such an advanced “Vibe Coding” setup. To help you optimize your config-layer-audit.md workflow, here is the recommended separation of concerns:

GEMINI.md (The “Who” and “How”): Use this for high-level personality, coding standards, and project-specific style rules. It defines the behavior of the agent within that workspace.
Skills (The “What”): Use Skills for discrete, repeatable capabilities or complex tool logic (like your Postgres or Redis integrations). Skills are best for specialized tasks that the agent should “call upon” only when needed.
\Doc Folders (The “Context”): Keep these for deep domain knowledge or API specs. Use the Knowledge Base feature to index these so the agent can RAG (Retrieve) them without cluttering the prompt instructions.

Best Practice: If an instruction applies to every prompt, put it in GEMINI.md. If it’s a specific “superpower” the agent needs to activate, make it a Skill.

1 Like

Working with the Knowledge Base is all new to me. Lookin at it it has a lot of obsolete information, so per Gemini recommendations, deleting everything and letting Antigravity background AI’s rebuild as needed. Exicited to see if it makes a difference :slight_smile: