If you have gpt plus plan

I burned my weekly limit in just 2 days. I was pissed off. That's why I created Lean Codex.

Lean Codex is a token-conscious workflow for OpenAI Codex. It reduces unnecessary context loading, tool calls, verification, and subagent delegation while preserving correctness.

It is inspired by Andrej Karpathy's observations about common LLM coding mistakes and by the multica-ai/andrej-karpathy-skills project. Lean Codex extends those ideas with explicit context, tool, verification, and multi-agent budgets.

It's very simple but for now it's working for me. And I'm constantly using it for creating custom extensions for java games. (That's why it consumes so much tokens). Now I see difference between 20-45% more efficiency, depending on the task. You can try it and tell me in the comments did it work for you.

Here is the git:https://github.com/heyseven7/lean-codex

8 comments

If you have gpt plus plan