Cubits and Queries

$ stop re-explaining yourself

Your AI forgets you every session. Here's the ten-minute fix.

Every tool starts from zero: ChatGPT, Claude, Cursor, all of them. Give them one shared memory and you stop briefing a blank slate every time. Here's the exact setup I use, minus my private notes.

The whole idea

One folder. One file. Every tool reads it before you type. That's the trick. The rest is just keeping it current.

Do it yourself, three prompts

Run Claude Code? Then you barely type anything. These three prompts build the memory, port it to your other tools, and keep it current. Not on Claude Code? They work the same in Cursor or Codex, just swap the filename.

1. 1

   Let it build the file for you

   Claude reads your whole repo, writes the memory file, then asks only for what it can't infer. One pass and you have a real CLAUDE.md.

   Paste into Claude CodeCopy

   Read this entire project, then create a CLAUDE.md at the repo root that a brand-new engineer, or a fresh AI session, could read to get fully up to speed. Infer the stack, architecture, key folders and the conventions the code actually uses. Then ask me only for what you can't infer from the code: who I am, what we're building and why, the decisions already made, and my hard rules. Keep it under 200 lines, add a Decisions section with newest first, and leave out anything the code already makes obvious.

2. 2

   Port it to every other tool

   One memory, every tool. This turns your CLAUDE.md into a Cursor and a ChatGPT version, and keeps all three in sync.

   Paste into Claude CodeCopy

   Using CLAUDE.md as the single source of truth, create a matching .cursorrules for Cursor and a CONTEXT.md I can paste into ChatGPT or Codex. Same facts, reformatted for each tool. From now on, whenever CLAUDE.md changes, update the other two to match.

3. 3

   Make it update itself

   End any working session with this and the file never rots. No stale notes, no 500-line dead doc nobody trusts.

   Paste at the end of a sessionCopy

   Before we wrap up: look back at what we changed and decided this session. Append any new decisions to CLAUDE.md, dated and newest first, update anything that is now out of date, delete lines that are no longer true, then show me the diff before saving.