Hotel competitive pricing analysis and travel planning system using Amap POI search, Fliggy hotel data, and multi-dimensional competitor scoring
27 tools with significant quality gaps. Most tools have basic descriptions (10-50 chars) but lack depth on WHEN to use them or prerequisites. Parameter descriptions are minimal or missing context. Input schemas present but sparse, many lack type constraints, enums, or ranges. No output schemas documented. Error handling absent. Tool names follow verb_noun pattern (good), but descriptions fail to distinguish similar tools (poi_search vs poi_around vs poi_detail). Composition issues: cache tools (get_cached_around_search, save_around_search, get_cache_stats) are low-level implementation details that should be hidden. Feishu push tools (push_text, push_card, push_daily_report) lack error guidance. No tool declares permissions or security implications.
添加一条价格记录到飞书多维表格
批量计算距离(本地计算,不消耗API配额)
清理所有过期缓存,返回清理的数量
驾车路径规划 → 返回距离和时间
对一批候选酒店进行过滤+评分,返回有效竞品列表(按得分降序)
地址→坐标转换(地理编码)
从缓存获取数据,如果有效返回JSON,否则返回None
Generic tool names ('get', 'set', 'stats') lack action clarity. LLMs cannot infer intent from 'get' alone, is it fetching cache, user data, or config? Rename to 'get_cache_data', 'set_cache_data', 'get_cache_stats'.
Descriptions are too brief (10-50 chars) and lack WHEN/WHY context. 'POI关键词搜索' tells WHAT but not WHEN to use poi_search vs poi_around. Add: 'Use when you have a keyword but no location. For nearby POIs, use poi_around instead.'
No output schemas documented. LLMs cannot plan downstream calls or extract fields. Document what each tool returns: poi_search returns List[POIRecord] with fields [id, name, location, address, distance_km, ...]. Add to every tool.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 46 | 2026-07-28+ | v2 |
获取缓存统计
获取缓存的周边搜索结果
按价格区间分组:low/mid/high,每个区间最多选10家,按得分降序
删除指定缓存
周边搜索(指定坐标半径内)带缓存:相同查询直接返回缓存,节省API配额
获取POI详情
POI关键词搜索(高德POI v5 API)
推送交互式卡片消息,美观格式
推送每日监测报告通知
推送文本消息到群/用户
查询酒店历史价格
坐标→地址转换(逆地理编码)
保存周边搜索结果到缓存
对单个竞品酒店进行多维评分
对单个竞品酒店进行多维评分(v2.1支持软性筛选)
批量搜索竞品信息
搜索酒店的价格和评分
写入缓存
获取缓存统计信息
天气预报(adcode: 120116=滨海新区)
Cache tools (get_cached_around_search, save_around_search, get_cache_stats, get, set, invalidate, clear_expired) are implementation details. Hide caching inside poi_around and other primary tools. Expose only high-level tools to agents.
No error handling guidance. Tools lack descriptions of failure modes and recovery steps. E.g., poi_search should document: 'Returns empty list if no POIs found. Try broader keywords or larger radius.' Feishu push tools should explain: 'Returns error if chat_id invalid, verify ID format (oc_* or ou_*).'