prime-radiant-inc/greenfield ? reverse-engineered prompt

Build me a Claude Code plugin called Greenfield that can take a codebase, docs, SDKs, runtime behavior, and even binaries, then reverse engineer them into clean behavioral specs that explain what the software does without exposing how the original code is structured.

I want a command like /analyze /path/to/target that runs a multi step review, gathers evidence from whatever sources are available, writes sanitized specs, test vectors, and acceptance criteria, and saves everything into a workspace folder with raw evidence, output specs, and a provenance trail for every claim.

It should also have a /sanitize /path/to/workspace command to clean up an existing workspace if the first pass left implementation details behind. Please make the output feel like something a fresh team could build from, and keep the specs focused on behavior, not code structure. Use current Claude Code plugin docs if you need to look anything up.