Can LLMs Predict Their Own Failures? Self-Awareness via Internal Circuits - View it on GitHub