Rules vs knowledge in CLAUDE.md | Context Engineering - courses.vizuara.ai

Rules or knowledge? The boundary is thinner than you'd expect. Here is the test that works most of the time: → How to do something → that is a rule → What to do, or what a thing looks like → that is knowledge Run a few lines through it. "As a chatbot, never say 'I don't know'." That is a rule. It constrains behaviour. "Retrieved context is injected with source attribution." That is knowledge. It is not telling the model how to act - it is describing what the system looks like. "Whenever there is any mention of refund, look at the refund policy document." Now it gets blurry. That is a rule and a piece of knowledge at the same time. "Start with cosine similarity." Closer to a rule - but you could argue for knowledge, and you would not be wrong. The point is not to win the classification argument every time. It is that you know which one you are writing, because the two behave very differently once they are in a CLAUDE.md file. Context Engineering - courses.vizuara.ai