What problem does it solve? CLAUDE.md is injected into every Claude Code session, but bloated or poorly structured files cause Claude to ignore instructions entirely. This Skill audits an existing CLAUDE.md against research-based context engineering criteria and produces a leaner, more effective version. ## Core Features & Use Cases - Instruction Auditing: Counts lines and imperative instructions, checking against the ~100-150 instruction budget that frontier LLMs can reliably follow. - WHAT/WHY/HOW Coverage Check: Verifies the file explains the tech stack, project purpose, and how to build, test, and verify changes. - Anti-Pattern Detection: Flags task-specific instructions, code style rules, stale code snippets, and hotfix-style directives that belong in hooks, linters, or separate docs. - Use Case: A team's CLAUDE.md has grown to 500 lines of accumulated rules. Run this Skill to identify what to remove, apply progressive disclosure by referencing existing docs, and generate a rewritten file under 300 lines. ## Quick Start Analyze my project's CLAUDE.md and rewrite it following best practices, reporting the results in Japanese.