For those running Claude Code on large monorepos , how are you structuring your CLAUDE.md without bloating context?
retained text
I've been using Claude Code heavily on a fairly large TypeScript/Go monorepo over the past few months. Early on, I made the mistake of dumping every coding standard, DB schema quirk, and test rule into a single root CLAUDE.md. Before long, it was 400+ lines and eating up tokens on every single turn—plus the agent started ignoring half the instructions anyway. Lately I trimmed the root CLAUDE.md down to around 80 lines (strictly build/test commands and high-level directory boundaries) and moved domain-specific rules into subdirectory CLAUDE.md files so they only get pulled in when the agent actually touches those folders. Curious how other senior devs here are organizing their project context or hooks for larger codebases? What's actually working for you in production vs. what turned out to be noise?
document fetched 2026-10-21 00:22 · canonical https://reddit.com/r/ClaudeAI/comments/1xalih0/for_those_running_claude_code_on_large_monorepos