Claude Code so bad it needs therapy? There's a doctor for that.

What it does claude-doctor is a TypeScript CLI that performs post-mortems on your Claude Code sessions. It scans ~/.claude/ transcripts for behavioral anti-patterns—structural, behavioral, and lexical—and generates ready-to-paste rules for CLAUDE.md or AGENTS.md. Think of it as a linter for your human-AI collaboration.

The interesting bit Rather than preaching generic best practices, it derives guardrails from your actual failures. It builds per-project signal baselines from your history, so the advice is personalized: if you constantly abandon sessions, thrash on the same file, or slowly devolve into one-word corrections, the tool notices and writes rules to address your specific dysfunction.

Key highlights

• Flags structural anti-patterns like edit-thrashing, error-loop, and read-to-edit ratios above 10:1
• Detects behavioral tics including correction-heavy user messages, keep-going-loop prompts, and negative-drift toward terse replies
• Scores sentiment with AFINN-165 plus custom agent tokens such as undo, revert, and broken
• Exports a persistent project model to .claude-doctor/ with model.json profiles and guidance.md for agent hooks

Verdict Useful if you maintain long-running Claude Code projects and want data-driven CLAUDE.md rules. If your AI chats are ephemeral one-offs, you won’t have enough history to bother.