26 tools with schemas visible in server.py, but critical gaps undermine quality. Descriptions are present but often verbose (200-400 chars) and written in Chinese, mixing implementation details with usage guidance. Parameter descriptions exist but lack constraint specifications (ranges, enums, formats). No output schemas documented. Error handling absent, tools return raw dicts with no recovery guidance. Tool names are action-oriented (remember, recall, link_memory) but some are domain-specific jargon ('consolidate', 'context_pack') that may confuse LLMs unfamiliar with memory systems. No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Composition is reasonable, tools are single-purpose, but chaining requires manual ID passing with no validation.

• Add JSON Schema output definitions for all tools. Document return type, field names, and types. Example: remember() returns {id: string, created_at: ISO8601, importance: float}.
• Rewrite descriptions in English, 50-150 chars, action-focused. Example: 'Store a fact or event with optional importance (0-1) and categories. Auto-estimates importance if omitted.' Avoid implementation jargon.
• Add enum constraints to parameters: importance [0.0, 1.0], priority [1, 2, 3, 4, 5], mode ['coding', 'research', 'chat']. Use JSON Schema minimum/maximum for numeric bounds.
• Implement error responses with recovery hints. Example: {error: 'memory_not_found', message: 'Memory ID abc123 not found.', suggestion: 'Call recall() to search by content instead.'}
• Add tool annotations: readOnlyHint=true for recall/get_memory/list_goals; destructiveHint=true for remember/consolidate/time_travel; idempotentHint=true for read-only tools.