My apologies! I left out the actual link in the post.

Here is the full protocol on GitHub: [https://github.com/DaibinThink/AI-Control-Protocol-Hardcore]

It's a structured system prompt designed to intercept those 9 failure modes. The core is using the Pre-Decision Engine (Chapter 3) to force the model to expose its blind spots before giving you an answer.

Would love to hear your thoughts if you test it out on local models or Claude.