review-claudemd

You can audit CLAUDE.md instructions by parsing recent JSONL conversation sessions from ~/.claude/projects to extract user and assistant text, then batching analysis prompts to identify violated rules and generate specific improvement suggestions for both global and local configuration files.

The best way to improve CLAUDE.md prompt quality is to review recent conversation patterns to detect where instructions are being ignored, then apply targeted suggestions to both global ~/.claude/CLAUDE.md and local project ./CLAUDE.md files based on observed behavior.

Yes, you can update global and project CLAUDE.md files separately. The audit process produces distinct GLOBAL suggestions for ~/.claude/CLAUDE.md and LOCAL suggestions for ./CLAUDE.md, allowing you to apply configuration changes independently based on conversation analysis results.

To safely back up CLAUDE.md before applying instruction updates, the workflow advises creating copies of both global ~/.claude/CLAUDE.md and local ./CLAUDE.md files prior to any edits, and supports drafting proposed changes for approval before they are committed.

CLAUDE.md instructions may be ignored due to ineffective or outdated rules. Analyzing recent JSONL session history helps detect violated instructions and outdated items, providing specific bullet-point improvements to strengthen your global and local configuration files.

Yes, CLAUDE.md conversation analysis works without external dependencies. It operates by parsing native JSONL session files stored locally in ~/.claude/projects, extracting conversation text, and aggregating findings into violated-instructions and suggestion reports.