Server has 6 tools with basic descriptions in Japanese, but significant quality gaps. All tools have descriptions (positive), but most are vague and lack actionable guidance. Input schemas are present but minimal, parameters lack type constraints, enums, and validation rules. No output schemas documented. Error messages are generic strings without recovery guidance. Tool names are reasonable (verb_noun pattern) but descriptions don't explain WHEN to use each tool or dependencies between them. write_dialogue and save_insight are near-duplicates, violating single-responsibility. No parameter descriptions beyond brief labels. Security concern: VAULT_ROOT path is hardcoded with placeholder, and no validation of relative_path prevents directory traversal attacks.

• Remove save_insight entirely; consolidate into write_dialogue. Reduce cognitive load on LLM tool selection.
• Add structured JSON output schemas for all tools. E.g., search_memos should return {results: [{path, score, snippet}], total_count, next_cursor}. Document in tool description.
• Validate and sanitize relative_path in read_memo and write_memo. Use pathlib.Path.resolve() and assert it stays within VAULT_ROOT. Reject paths with '..' or absolute paths.
• Add parameter descriptions with constraints: query (1-500 chars, natural language), limit (1-100, default 10), folder (enum: ai_dialogues, 10_chatgpt_dialogues, 11_claude_dialogues, or custom).
• Enhance error messages with recovery guidance. E.g., 'Index not found. Run index_vault.py to create it, then retry.' or 'File not found. Try search_memos() to find similar notes first.'