Your Claude Code rules are a liability you'll never audit

If you've been using Claude Code for more than three months, your .claude/rules/ directory is full of instructions written for yesterday's model. Newer frontier models handle most of those defaults correctly on their own — but the old rules are still there, occupying context on every message, sometimes fighting the model's improved defaults. Nobody is auditing them because nobody agreed on what "auditing a prompt" even means. This post proposes one answer: every persistent rule carries a WHY tag