HMA server has well-structured schemas and detailed descriptions, but suffers from critical naming and composition issues. Tool names lack clear action verbs (memory_write, memory_query are acceptable but memory_query_anchors and memory_resolve are vague). Descriptions are lengthy (200-400+ chars) and domain-specific but often bury actionable guidance. Parameters are heavily documented with constraints, but many descriptions are in Chinese, reducing LLM accessibility. No output schemas are documented. Error handling is mentioned in descriptions but not formalized. The server exposes complex internal concepts (QueryEnvelope, ENVELOPE_VIOLATION, corpus_missing_entity) that should be abstracted. Five tools operate on overlapping concerns (query, query_anchors, resolve) without clear composition guidance.

• Document output schemas for all tools. Example: memory_query should return {results: [{id, title, summary, score}], total_count, next_cursor}. This enables LLMs to chain tools and extract the right fields.
• Consolidate query tools: merge memory_query, memory_query_anchors, and memory_resolve into a single search_memory tool with a 'search_level' parameter (package|anchor|with_disambiguation). Reduce cognitive load and eliminate composition ambiguity.
• Translate all descriptions and parameter docs to English. Use clear, LLM-friendly language: 'Search for memory packages by keyword' instead of domain jargon. Remove internal implementation details (BM25, linked-BFS, QueryEnvelope).
• Add enum constraints with descriptions for 'mode' parameter: {single: 'Single question, no decomposition', multi: 'Multi-step query with sub-questions', enumerate: 'Range enumeration'}. Help LLMs classify queries correctly.
• Add error handling guidance to descriptions. Example: 'If no results found, try broadening keywords or removing scope filter. If corpus_missing_entity error, the entity is not in the knowledge base, suggest alternative topics to the user.'