Asked Claude what in my docs was hurting its reasoning. The answer changed how I think about CLAUDE.md files

Context: ~3 weeks building a product almost entirely with agents. I was disciplined about docs the whole time. Milestones checked off as they shipped, design decisions logged, gotchas recorded. If anything I over-documented. Then on a whim I asked Claude to "evaluate what might be hurting agents' thinking" in the repo docs. Some of what it found: - The three docs on my read-before-you-code list totaled 1,085 lines, and roughly 75% was finished work. Fully-checked checklists, phase completion logs, old verif