Claude Code Quota Blowing Up Early Again? Someone Used a Prompt to Find the Three Biggest Token-Burning Pitfalls

Claude Code users may have had the same feeling recently: the usage limit is becoming less and less durable. You clearly haven’t done much work, yet “Usage limit reached” still shows up right on time.

Developer Khairallah AL-Awady shared a method on X: paste a prompt into Claude Code, and it will automatically go through your session records from the last 7 days, identify the three biggest token leaks, and provide corresponding rules to plug them. It won’t change anything directly; it will wait for your approval first.

Here’s roughly what the prompt does:

• Build a quick script that pulls token counts for the past week from ~/.claude/projects; the log files themselves are not loaded into context, to avoid burning more tokens the more it investigates.
• Rank the three most wasteful token usage patterns, such as opening an entire file, pasting huge test/build output, launching subagents, or leaving a session hanging for hours. Show roughly what percentage each one consumes, plus a rule that could have prevented it.
• Only after you approve will it write the rules into the project instruction file. Prefer CLAUDE.md; if it doesn’t exist, create AGENTS.md. Secrets are automatically masked.

This method isn’t limited to Claude Code. If you use Grok Bot, send it this prompt as-is and it will execute it on your machine. If you use Codex, change the path in the prompt to ~/.codex/sessions and write the rules to AGENTS.md.

A bit of background: CLAUDE.md is Claude Code’s project-level instruction file, essentially the project’s “usage guidelines,” telling the model what it can and cannot do. AGENTS.md is the corresponding file for other AI coding tools (such as Codex). The value of this prompt is that it lets the AI analyze its own behavior and then write the rules into the file.

Tools are getting smarter, but your quota and patience are still limited editions. If you also often get blocked by the quota, you can run this prompt first and see who is really eating your tokens behind the scenes.