Claude Audit

Read and audit all CLAUDE.md files in the system.

Step 1: Discover All CLAUDE.md Files

# Global
cat ~/.claude/CLAUDE.md

# Global rules
ls ~/.claude/rules/

# Project-level
find . -name "CLAUDE.md" -not -path "*/node_modules/*" 2>/dev/null

# Memory
cat ~/.claude/projects/*/memory/MEMORY.md 2>/dev/null

Step 2: Read All Files

Read each discovered file completely.

Step 3: Audit Each File

Check for Redundant Instructions

• Instructions that duplicate default Claude behavior
• Instructions repeated across multiple files
• Instructions that contradict each other

Check for Verbosity

• Long explanations that could be one line
• Examples that aren't necessary
• Sections that could be merged

Check for Misplaced Content

• Content that belongs in a skill instead of CLAUDE.md
• Content that belongs in MEMORY.md instead of CLAUDE.md
• Project-specific content in global files

Check for Stale Content

• References to deprecated tools or APIs
• Instructions for workflows no longer used
• Outdated model names or endpoints

Step 4: Output Report

## [file-path]

### Redundant Instructions
- Line X: "[instruction]" → Already default behavior / duplicates [other file]

### Verbose Sections
- Section "[name]": Could be condensed from N lines to: "[suggested replacement]"

### Misplaced Content
- "[content]" → Should be in [skill/memory/other file]

### Stale Content
- "[content]" → Outdated because [reason]

### Suggested Rewrite
[condensed version of the file]

Step 5: Summary

• Total files audited
• Estimated token savings if suggestions applied
• Priority order for cleanup
• Specific merge/move recommendations