Hi everyone ![]()
I wanted to share something I’ve been building with Antigravity 2.0 that I think might be useful to other vibe coders here.
Quick backstory: I coded professionally 20+ years ago, then stopped. I recently came back to build a full-stack app — but this time I don’t write code. I command it. Antigravity is my IDE, Gemini and Claude are my engineering team, and I just set the direction.
Along the way, I developed a workflow infrastructure that I’ve extracted into a reusable template. The key innovation is meta-prompting — prompts that analyze your live codebase and generate context-aware review prompts automatically.
What’s in the template
-
Meta-prompts (
docs/prompts/_meta/) that scan yourpackage.jsonfiles,GEMINI.mdrules,.skills/directories, and MCP server config to generate hyper-specific code review, spring cleaning, and architecture review prompts -
Modular skills (
.skills/) — domain-specific runbooks the AI loads lazily instead of bloating every conversation’s context -
Slash-command workflows (
.agent/workflows/) —/code-review,/spring-cleaning,/architecture-review,/run-meta-prompts, and more -
Report retention — timestamped reports with archival policies so the AI reads its own history and doesn’t re-flag resolved issues
-
Dual-model implementation review — Gemini reviews the plan as architect, Claude stress-tests it as pragmatist
-
Gemini Gems integration — a
consolidate_docs.ps1script that merges all.mdfiles so you can upload them to a Gem and ask your repo questions from the Google Gemini app on your phone
The repo
GitHub: https://github.com/ThMoJe/antigravity-vibe-template
Fair warning: I haven’t published anything to GitHub since before Git existed (I’m a CVS/SVN era survivor
). If I’ve committed some open-source faux pas, I’d genuinely appreciate a heads-up.
This template was extracted from a production app I’m building entirely with Antigravity, and I’m close to releasing it. I’m sharing because I think my approach — as a non-developer doing vibe coding — might offer a different perspective than what experienced developers would build. Sometimes not knowing “the right way” leads to interesting solutions.
Feedback, suggestions, and PRs welcome!