Repo archaeology for the vibe-coding era

What it does

GitReverse is a web app that ingests a public GitHub repository—its metadata, root-level file tree, and README—and asks an LLM to write a short, conversational prompt. The idea is that you could paste that output into Cursor, Claude Code, or Codex and “vibe code” the project from scratch. It is largely glue between the GitHub API and whichever LLM key you provide.

The interesting bit

The app treats the repository as a finished product and the prompt as its source code, flipping the usual development flow. Shareable /owner/repo paths let you link directly to a reverse-engineered recipe, and an optional Supabase backend can cache results so popular repos do not need to be re-analyzed.

Key highlights

• Supports four LLM providers out of the box: Grok, OpenRouter, OpenAI, and Google AI Studio, with an auto fallback chain.
• Optional Supabase integration adds a browsable /library and per-user /history.
• Deep and focus modes can delegate to a custom backend service for more detailed analysis.
• Built on Next.js 16 App Router and React 19.

Caveats

• Subfolder-aware context is not implemented yet; /tree/... URLs redirect to the repository root, so the prompt is always scoped to the whole project.
• The analysis is shallow by design: only depth-1 file tree, README, and metadata feed the LLM, so nuance from deep source files is lost.

Verdict

Useful if you want a high-level recipe for vibe-coding a clone, but not if you need subfolder-level precision or a true structural decompiler.