Hook

Their other posts in the index, biggest breakout first.
You need to steal this tip from the creator of Claude Code if you are building in 2025. The guy who built Claude Code, Boris Cherny, recently shared a strategy for his Claude.md file on X, and his workflow is so good it's backed by Stanford researchers. If you don't know what a Claude.md file is, it's basically a persistent, high-level set of instructions for your Claude Code agents. This file is extremely important as it guides your AI's behavior, informing it of things like project structure or patterns, thus reducing errors and saving you money. This is a markdown file based on his post, but the real genius of it comes from a single instruction: the self-improvement loop. This is a clause in the markdown file that informs the AI that after any code correction, to update another markdown file with the lesson it learned. This allows Claude Code to write rules for itself. Thus, the longer you use it, the smarter it gets on your code base. It's actually a studied phenomenon for researchers at Stanford and UC Berkeley called agentic context engineering. Results from the study found that evolving contexts enable scalable, efficient, and self-improving LLM systems with low overhead. If you want to give this tip a try for yourself, comment "markdown" and I will DM you the file!