Keep one project brief every AI tool can read

Switching tools at every limit means re-explaining the project · X · @aditiitwt · Oct 2026

Rotating between Codex, Claude Code and Cursor to dodge limits meant repeating the same project explanation in each one.

Lesson

Keep one plain project notes file (stack, where things live, what not to touch) in the repo and point every tool at it.

Fix prompt

Copy this into your own AI tool (Cursor, Lovable, Claude Code, Bolt…). It's general guidance written by us. Review what the AI suggests before applying it.

I keep losing context with Codex (it forgets rules, re-asks about my project, or a model switch undoes work). 1) Create a short project brief file for the repo (e.g. AGENTS.md or CLAUDE.md): stack, how to run it, key files, files never to touch, and coding rules. 2) Tell me which rules should be enforced by tools instead of prompts (git hooks, linters, tests, ignore files) and set those up after I approve. 3) Remind me to commit working steps with clear messages, and show me how to roll back. Keep the brief under 60 lines.

Source: X ↗