Loading ToolBench
Vietnamese Wikipedia knowledge graph QA system with Neo4j backend. Provides tools for searching, exploring entities, finding paths, and answering complex multi-hop questions using graph-based retrieval and LLM synthesis.
WikiGraphRAG has 7 well-named tools with clear, contextual descriptions (avg 150 chars). All tools have input schemas with types and defaults. However, output schemas are not documented in the source code provided, and error handling guidance is absent. Parameter descriptions are present but lack constraint details (ranges, enums, formats). Tool composition is sound, each tool has a single responsibility and names follow verb_noun convention. The descriptions include helpful usage guidance ('Do NOT use for...') which aids LLM selection. Missing: documented return types, error recovery hints, and validation rules for parameters like top_k and depth.
Use for complex questions requiring reasoning across multiple facts. Automatically decomposes multi-hop questions, retrieves from multiple sources, and synthesizes an answer with citations. Do NOT use for simple factual lookups — use search_knowledge_base instead.
Use when you know an entity name and want to see what connects to it. Examples: explore 'Hồ Chí Minh' to find related people, places, events. Do NOT use for searching — use search_knowledge_base first to find entity names.
Use for questions asking HOW two things are connected or related. Examples: 'Mối quan hệ giữa Phạm Văn Đồng và Hồ Chí Minh?' Requires knowing both entity names — use search_knowledge_base first if unsure.
Use for broad topic overviews before diving into specifics. Returns a summary of related concepts clustered around a topic. Examples: get overview of 'Chiến tranh Việt Nam' or 'Văn học Việt Nam'. Do NOT use for specific factual questions — use search_knowledge_base instead.
Get statistics about the knowledge graph. Returns counts of pages, chunks, entities, and relationships. Useful for understanding the scope and coverage of the knowledge base.
Output schemas not documented. LLMs cannot plan downstream calls or extract fields without knowing what each tool returns. All 7 tools lack return type documentation.
Parameter constraints missing. top_k accepts 1-20 but no min/max in schema; depth accepts 1-2 but no enum; max_hops defaults to 5 with max 6 but no bounds declared. LLMs may pass invalid values.
No error handling guidance. Tools lack recovery hints (e.g., 'If entity not found, try search_knowledge_base first'). Agents cannot self-correct on failures.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 68 | <=2025-11-25 | v2 |
List all entity types in the knowledge graph with counts. Returns entity types (Person, Organization, Location, Work) and their counts. Useful for understanding what kinds of entities are available.
Use for factual questions needing 1-2 facts from Vietnamese Wikipedia. Examples: 'Ai sáng lập Đảng Cộng sản Việt Nam?', 'Hà Nội có bao nhiêu quận?' Do NOT use for questions requiring reasoning across multiple facts. Returns ranked text passages with source page links.
Parameter descriptions lack format/range details. 'entity_name' says 'Vietnamese' but no guidance on case sensitivity, diacritics, or partial matches. 'question' parameter has no length limits or language constraints stated.
No pagination support documented. search_knowledge_base returns 'ranked text passages' but no indication of total count, next_cursor, or how to fetch additional results beyond top_k.