Post

Using a different model family as the reviewer is the smart part. A reviewer from the same family as the author tends to share its blind spots. Has Argon caught anything on a diff that Claude Code or Codex had already signed off on? That's the comparison I'd want to see. 🔮

I haven't run the same rubber duck agent with a model of the same family to compare, but that was after several iterations of validation from Claude. The model matters, but the harness as well, the Copilot rubber duck agent is excellent at adversarial reviews with another model

Cross-family review has caught real bugs for me too. One tweak that helped: give the reviewer only the diff and the spec, not the author's chat, so it can't inherit the reasoning.

Do you use Claude Code, Codex or Cursor, but want to leverage the rubber duck reviews from GitHub Copilot? I've got you covered! github.com/awakecoding/...

The rubber duck bridge is actually written in Powershell too. github.com/awakecoding/...