A CLAUDE.md file is a plain document Claude reads automatically at the start of your work — your role, preferences, and rules. Set it once and every conversation starts already briefed.
Without a memory file, every new conversation starts from zero. You re-explain who you are, how you want things written, what your project is about — and you get a good answer, but the setup evaporates the moment the chat ends. Multiply that across a week and it's real time lost.
The fix isn't a better prompt each time. It's writing the briefing down once in a place Claude reads on its own.
A CLAUDE.md is just a plain text/markdown file that Claude reads automatically when it starts working in that space. You don't paste it or trigger it — it's always-on context. Think of it as the standing orders you'd hand a new assistant so they don't have to ask the basics.
It lives with your project (or your workspace), so the right context shows up for the right work.
Keep it to what actually changes Claude's behavior: who you are and your role, the tone and format you want, your goals, and clear always/never rules. Add where things live and what 'done' looks like for your common tasks. Skip anything that doesn't change the output.
Shorter and sharper beats long and vague — this is a briefing, not a manual.
# About me Army officer, non-developer. Plain English only. # How I work Concise. No jargon. Show me the steps. # Always / Never Always: explain the 'why' in one line. Never: use the words 'leverage' or 'delve'.
The file gets more valuable the more you tend it. Every time you correct Claude — 'use metric units,' 'don't add a summary at the end' — add that as a line. Every so often, trim what's gone stale. Over a few weeks it compounds into an assistant that already knows your preferences.
Each correction you write down is one you never have to repeat.
A fill-in-the-blank memory file so Claude starts every conversation already briefed on who you are and how you work.