Hitting the same wall every day across Claude Code, Codex, and Cursor and want to know how the rest of you are handling it.
Open a new session on a project I worked on yesterday → first 2-4 minutes the agent is grepping around rediscovering what files exist, the architecture. Most things I figured out in yesterday's session are gone, until I ask it to save that as well. Switch from Claude Code to Codex mid-task and the whole rebuild happens again, neither tool knows what the other just learned.
I've been maintaining CLAUDE.md and AGENTS.md but they mostly capture static rules ("we use snake_case"), and not narrative on what am I building why am I building and all. And they rot, I am not updating docs in the middle of coding.
Curious what's actually working for you:
Do you maintain MD files by hand? Like a LLM wiki in the repo.
Anyone running a memory MCP server? Does it actually work or is it one more thing to babysit?
For people switching between Claude Code, Codex, and Cursor — how do you keep them in sync, if at all?
Or have you just accepted the friction?
Top comments (0)