Got a little prompt on hand recently for whenever Claude Code starts bumping into the usage limit. Paste this in:
Don't change anything without my approval. Write a script that goes through my Claude Code session logs under ~/.claude/projects from the past 7 days. Use the token counts already stored in those logs and don't load the raw files into your context. From the numbers, identify the 3 habits that cost me the most tokens, e.g. reading entire files, pasting long test or build output, spawning subagents, or letting one session run for hours. Estimate what share of my usage each one accounts for, and draft a single rule that would cut it. Once I confirm, add the rules to the instruction file this project actually loads: CLAUDE.md if there is one, otherwise AGENTS.md. Never print secrets.
It reads the last 7 days of session history (only the token counts, so nothing big gets pulled into context), calls out your 3 most expensive habits with a rough % for each, and writes one fix per habit. Nothing lands in your files until you say yes.
Same trick works on Codex: swap the name, use ~/.codex/sessions as the path, and have the rules go into AGENTS.md only.
@jeffbruchado Live chat, either by you or your AI double who has learned your question and mindset. But you can definitely book a call with user in chat for sure !
This is the signal that you took a good care of your codebase - diff is only ~+1000 lines for a major change.
If I let Claude/Codex drive autonomously without review, I think the diff could be at least +5000 lines.
Building a small app is one thing.
Getting people to discover it, trust it, and actually buy it is a completely different challenge.
For indie developers: which part has been hardest for you?
@sagarmoy22@X Heyyy๐ Let's connect! I'm building FounderPing - it lets you meet your users right in your app, and now it supports web and Android. ๐
@Deeprenuer Hi there! ๐ I'm building FounderPingโit lets you meet your users right in your app, and now supports web and Android. I'm also a dog father!