Diagnose your Claude Code harness (CLAUDE.md, settings.json, .mcp.json, hooks, skills, memory, MCP servers, and workflow) against 25 anti-pattern indicators. Outputs an S–E overall grade, per-category scores, violations with evidence citations, and prioritized Quick Wins. Includes an interactive fix flow: `--fix-mode propose` emits structured JSON proposals that Claude Code applies via the Edit tool with per-item user approval. Deterministic (no LLM — grep/AST/JSON only), zero runtime dependencies, bilingual (Japanese default, English via `--lang en`). When installed as a plugin, the bundled skill auto-triggers on natural-language requests like "diagnose my harness" and exposes `/neko-harness-doctor` as a slash command.