marco-polo
马可波罗:单城市旅行规划。用于需求与交通比较、住宿和景点筛选、预约准备、高德空间路线与预算,以及独立美食攻略、城市名片和交互旅行手册,也用于继续或修改已有行程。
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Badge for your README
Markdown
[](/listings/megumi-ben--marco-polo-skill)HTML
<a href="/listings/megumi-ben--marco-polo-skill"><img src="/api/listings/megumi-ben--marco-polo-skill/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
1. Install Marco Polo
1. **[Download the Skill ZIP](https://megumi-ben.github.io/marco-polo.skill/downloads/marco-polo.zip)**. 2. Extract it and place the `marco-polo` folder in your project's `.agents/skills/` directory.
The entry point should be `.agents/skills/marco-polo/SKILL.md`. The package contains only the skill instructions, agent metadata, output specification, and map template—no demo website, screenshots, deployment files, or Git history.
2. Connect travel research and maps
Install **the Xiaohongshu skill bundle and both Amap skills**, then configure your own Amap credentials and Xiaohongshu login. Add **FlyAI** when you need transport or specific hotel searches. Place these skills alongside `marco-polo/`; see [Dependencies](#dependencies) for download links and setup.
Your agent also needs web research access to check official travel information.
3. Describe the trip you have in mind
For example:
$marco-polo I'd like two relaxed days in Beijing, with historic architecture and parks. I'll arrange travel to and from the city myself, but haven't chosen where to stay. Help me plan.
Start with what you know. Marco Polo asks for the dates, group size, budget, and other information needed for the current stage. Later, you can continue with requests such as:
> My hotel has changed to this address. Update the first and last journeys each day, along with their costs.
<details> <summary>Other locations, updates, and output folders</summary>
Use the **Skill ZIP** above for installation. GitHub’s **Code → Download ZIP** and `git clone` provide the full development repository, including the demo website; see [Contributing](CONTRIBUTING.md) if you want to work on the project.
- For personal use across projects, you can install to `~/.agents/skills/marco-polo/`. See the [official Codex skill locations](https://learn.chatgpt.com/docs/build-skills). If the skill does not appear, start a new session or restart Codex. - To update, download a fresh Skill ZIP. Move your existing skill folder outside the skill directories as a backup, then extract the new package. Unzipping over the old folder would leave old demo files behind. Preserve any local configuration you have added. - Results are created progressively in `travel_plan/<city-trip-id>/` under your working directory. Revisions reuse the same data, so you can continue where you left off.
</details>
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
SKILL.md · Manifest · 150 lines
View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)
- ---
- name: marco-polo
- description: 马可波罗:单城市旅行规划。用于需求与交通比较、住宿和景点筛选、预约准备、高德空间路线与预算,以及独立美食攻略、城市名片和交互旅行手册,也用于继续或修改已有行程。
- ---
- # 马可波罗:单城市旅行规划
- 让用户知道为什么住这里、为什么去这些地方、为什么这样走,并拿到能照着执行的行程。保留实际使用中积累的约束,把流程和文件理清楚,避免为了完整而增加无用工作。
- ## 范围与执行原则
- - 默认一次规划一个城市,以一个住宿基地为主,可包含近郊一日游。多城市请求先明确本轮范围或分城处理,保留已定城际交通边界;不要遗漏城市后声称整趟旅行已完成。
- - 接受自然语言,不要求用户填 JSON。继续或修改时先读已有行程数据和状态;已给出的信息不重复问,已买好的交通不重新比较。
- - 五阶段:**①需求与可选大交通 → ②住宿、③景点可并行研究 → ④路线规划、美食研究、城市名片素材三线并行 → ⑤分别制作页面,统一入口交付**。路线分支保留资料与预约前置、高德空间排程和真实交通校验;并行指工作之间无需串行等待,不要求使用 subagent。
- - 住宿和入选景点未确认前,不生成详细每日路线;路线未确认前,不交付定稿地图。用户明确授权代选时,记录采用的选择即可,不逐项重复询问。必要假设和未核验项要可见。
- - 美食与城市名片独立于每日路线:资料就绪即可制作各自页面,无需等待路线确认,也不要求用户逐店确认。默认交付三个独立页面与统一入口;用户明确只需部分成果时遵从其范围。
- - 每次产出决策文件,都在对话中给出摘要、默认推荐、理由和少量待决策项,不能只让用户翻文件。
- - 收到用户选择或修改后,先记录到数据与状态,再开展后续查询;每个分支完成即保存成果,不等其他分支或网页全部完成才落盘。
- - 小红书用于经验与口碑,官方来源用于开放、票务、预约等事实。商业查询价格记录查询时间与适用日期,估算明确标注;查不到不能当作免费、无需预约或已经核验。
- ## 工具与文件
- 按需读取依赖技能,查到结果后继续本阶段,不把一次检索完成当作整趟规划结束:
- | 能力 | 用途 |
- |---|---|
- | `$flyai` | 需要时比较机票、火车票、酒店与旅行产品 |
- | `$xiaohongshu-skills` | 住宿、景点、美食经验;使用其登录与搜索流程 |
- | `$amap-lbs-skill` | 高德 POI、坐标、入口与步行/公交/驾车路线 |
- | `$amap-jsapi-skill` | 最终地图展示与交互 |
- | 网络搜索及官方来源 | 核验出行日期适用的营业、票务、预约、交通与天气信息 |
- 依赖的下载与安装见 [在线安装说明](https://github.com/megumi-ben/marco-polo.skill/blob/main/README.zh-CN.md#依赖安装),只在配置环境时阅读;本包不附带依赖源码或小红书操作指南。
- 所需能力不可用时说明缺口,继续不依赖它的工作。没有取得小红书内容不能声称已做双源研究;没有可靠坐标或路段数据不能声称路线已经验证。
- 首次写入或接续旧行程时读 [输出规范](references/outputs.md)。成果放在 `travel_plan/<城市-行程标识>/`,按阶段创建,不先造空文件。结构化事实保存在 `行程数据.json`,金额由 `费用跟踪.csv` 维护,可读文档和各页面使用这些数据。
- ## ① 收集需求与可选大交通
- **输入:** 用户描述、已有文件和预订。
- 1. 提取城市、日期或天数、人数及成员限制、预算口径、偏好与强度、必去/不去、已有住宿和固定活动;记录已知饮食忌口、必吃项目与餐厅预约,缺少一般口味偏好不阻塞研究。预算默认人均,说明是否包含大交通与住宿;预算、兴趣暂缺不阻塞城市研究,用户明确的预算上限和限制必须遵守。
- 2. 区分出发日、实际到达、住宿晚数和离开时间,不能把路上过夜算成目的地游玩。只有天数时可以先研究;详细排程前补齐日期和时间边界,或明确以用户接受的假设做草案。
- 3. 确认大交通是已购买、需辅助规划还是自行安排。需要比较才补问出发地、返程地并调用 FlyAI;比较门到门耗时、价格、换乘、第一天/最后一天可用时间,必要时比较周边机场或高铁枢纽。跳过规划仍记录到离时间和站点。
- 4. 初始化需求、状态与费用。接受推荐不代表实际购票或预订。
- **输出:** `行程数据.json`、`行程状态.md`、`费用跟踪.csv`;需要交通比较时生成 `大交通方案.md`。
- **交接:** 城市和基本时间范围清楚,可开始②③;只补问影响当前工作的缺失信息。
- ## ② 推荐住宿区域
- **输入:** 需求、预算、交通节点;可参考③的粗略景点分布,无需等待用户选完景点。
- - 网络与小红书两路研究,推荐 1—3 个区域。按交通、主要景点分布、餐饮、安静程度、预算和同行者需求给 10 分制推荐分,并解释优缺点与适用条件。
- - 默认只选区域,不提前展开酒店榜单;用户要求具体酒店时再用 FlyAI,已有酒店则直接记录真实住宿。
- - 景点筛选后复核住宿是否匹配,明显不匹配时给出取舍,不擅自更换已订酒店。
- **输出:** `住宿区域候选.md`;把选中的区域或酒店写回数据和状态,区分区域参考点与真实酒店。
- **交接:** 用户选择已明确。②③可一起汇报,也可分别确认。
- ## ③ 推荐并筛选景点
- **输入:** 城市、时间、偏好、预算和刚性要求;住宿位置只作参考,不据此漏掉值得去的远处项目。
- - 常规搜索与小红书共同收集景点、街区、美食区域和体验,复用已有攻略。以细读 5—10 篇相关笔记为目标,识别过时、重复与争议信息;取得不足时说明原因,不伪造共识。
- - **先扩展候选,再由用户筛选收敛。** “每天 2—4 个”等强度要求约束最终排程,不是初始候选数量。按城市供给与用户偏好覆盖不同类型、片区和有意义的替代项,留出实质性的比较空间;通常多于最终入选数量,不提前裁成刚好可排满的组合。最多 30 个,不凑数;合适选择有限时说明。
- - 每个候选给出类型、片区、推荐理由、优先级、停留时长、适合时段、基础门票/预约信息和明显限制。
- - 候选阶段先查会改变选择的信息:是否收费、预约难度、已知闭馆或停业等。紧迫放票事项立即提醒,不等全部筛选结束;其余详细核验集中在入选后。
- - ②③研究时顺手保存美食线索、景点简介与图片来源,供④复用;候选阶段不必为全部地点补齐照片,用户入选后再深入整理。
- - P0 强烈推荐、P1 推荐、P2 有时间再去、P3 替代备选;不必每级都有。模型推荐与用户选择分开,用户必去另记。对话中按类型或优先级呈现候选概览,突出最推荐项目、差异和争议;重点推荐不能代替其余候选的展示,方便用户比较后取舍。
- **输出:** `景点清单.md`,持续记录候选、用户选中、备选和排除项。确认后在顶部汇总最终名单,不另复制一份已选清单。
- **交接:** 入选集合明确;可立即开展④A,正式排程仍需住宿明确。
- ## ④ 路线规划、美食研究与城市名片并行
- **输入:** 入选与备选地点、需求、已有预约及②③研究结果;B/C 正式排程还需住宿选择和明确的时间边界。
- 以下 A→B→C 是路线主线;D 美食与 E 城市名片素材可独立推进,不等待每日分配或路线确认。路线确认只约束地图制作;各内容分支完成即可进入⑤制作对应页面。
- ### A. 入选资料与预约前置
- - 用官方或可靠来源核验入选项目在出行日期的门票、开放、闭馆、停止入场、预约规则;分别记录建议游玩时段和停留时长,补充到 `景点清单.md`。
- - 有预约事项就先生成 `景点预约.md`,按需要行动的时间排序。预约规则和放票窗口先查;目标日期、时段与路线联合确定;已经成功预约的时段作为固定约束。无需等住宿确认才研究预约。
- - 未放票允许做带前提的草案,预约失败要提供换日、换时段或替代项目。用户采用路线不代表已约成功;规划不自动执行下单或预约。
- ### B. 高德空间规划:每日路线的必要前提
- 1. **先查位置。** 用高德 LBS 获取住宿、入选地点、必要车站与用户固定活动地点的 POI、地址、经纬度和匹配结果。核对同名点、大型景区入口及同一景区内部点;不盲信导入的 KMZ 或名称推断位置。记录坐标系和“经度、纬度”顺序,不混用不同坐标来源。
- 2. **再理解分布。** 依据坐标识别片区和远近关系,考虑河流、道路、围墙与入口绕行。按地理簇分天,并结合预约、闭馆和适合时段调整;不能只按热度排名或机械最近邻排路线。
- 3. **形成日程草案。** 通常从住宿出发并返回;抵达、离开日使用实际交通节点,加入行李与进出站时间。到达日下午可合理利用;离开日倒推交通与候车余量,不能安排赶不上的夜游,可与其他空闲夜晚互换。
- 4. **用实际交通校正顺序。** 为草案中的相邻地点查询适用路线,必要时比较不同顺序。直线距离仅作初筛,不能代替道路距离和耗时;费用无可靠结果时使用当地依据估算并注明。记录查询时间,未来日期不把当前交通结果说成当日保证。
- 沿用以下查询起点,并根据道路、体力、行李和天气调整:
- - 不超过 800 米:优先查 walking,校正步行道路距离与时间。
- - 800—1500 米:先查 walking;绕行、疲劳、天气或赶预约时补查打车/公共交通。
- - 超过 1500 米:比较步行、驾车与公共交通;接口不适用或步行明显无意义时说明,不机械凑齐结果。
- - 不查询骑行。累计步行和景区内部步行也计入体力;无法取得可靠路线时标注未核验,不拿直线距离伪装已验证交通。
- ### C. 把路线排成能执行的一天
- - 景点、交通、用餐、休息一起计时;保留排队、取行李和赶车余量。普通用餐保留时间、区域与预算,不等待具体餐厅推荐、不将候选店铺自动排入行程。用户选中的美食街和夜市仍按景点排程;明确必吃或已预约餐厅作为固定需求处理,不能因美食分支独立而漏掉。
- - 模型给出默认交通推荐。仅把明显取舍交给用户:例如耗时差小而费用差大、较长步行但体验好、行李/赶预约/高温等特殊限制,或地图结果与现场可达性冲突。
- - 检查开放与预约冲突、首尾日边界、总耗时、累计体力、预算及备选是否可用。同一门票、联票或景区内部点不能重复计费、重复计算整段游览时间。
- - 入选项目必须有安排或明确解释。放不下时说明冲突和少量取舍,不悄悄删掉用户必去项目,也不靠不合理压缩停留时长凑数。
- **路线输出:** `每日行程.md`,包含按天路线、安排理由、相邻交通、用餐时段与区域、休息、当天预约摘要、替代项、费用与检查结果;更新数据、预约和费用文件。按天/路段写短段落,避免超宽表;复杂交通比较才另建专题文件。
- **路线交接:** 用户确认路线;待预约、暂定和未核验项明确保留。关键位置或交通不足、存在未解决硬冲突时,保持草案,不声称可执行性已验证。核心条件变化回到相关阶段,普通路段调整在本阶段完成。
- ### D. 独立美食攻略
- - 复用②③线索,继续用网络与小红书搜索本地菜、小吃、饮品和具体店家,读取相关笔记正文,区分亲历体验、推广、争议与过时信息。没有取得小红书内容时明确研究缺口,继续可用来源,不伪造口碑。
- - 分清“吃什么”和“去哪吃”:菜品给推荐理由、口味与忌口提示;店家给招牌、具体分店、区域、地址、人均依据、营业时间、排队/预约情况与替代选择。用高德核对门店与位置,动态信息保留来源和查询时间,查不到标待核验。
- - 按区域或类型组织,让用户自行选择;不绑定具体日期,不要求先确认路线,不把店家自动加进日程或地图。只记录候选参考价,不将全部推荐店铺入账;路线中的餐饮预算仍正常维护。
- **输出与交接:** `美食推荐.md`,结构化事实与来源写回共享数据;资料就绪即可制作 `food.html`。无需逐店确认;仅在必吃或已有预约影响行程时同步路线约束。
- ### E. 城市名片素材
- - 简述城市特色,围绕已入选景点整理看点、游玩提示与对应图片;复用③及 A 的资料,不重做候选筛选,不另建重复的城市介绍、坐标或图片清单。备选若展示必须与入选项区分。
- - 查 POI 时尽量同次取得坐标与照片(适用时使用 v3 place/text 的 extensions=all)。保存来源、匹配级别与可用 HTTPS 链接;片区图与菜品/门店图说明对应关系,缺图待补,不用相似地点或菜品冒充,不阻塞路线。
- **输出与交接:** 补充 `景点清单.md` 和共享数据中的城市简介、景点看点与图片记录;无需每日顺序,资料就绪即可制作 `city-impression.html`。
- ## ⑤ 分别制作页面,统一入口交付
- **输入:** 各分支已整理的资料;地图额外要求已确认的日程、地点与交通数据,以及最新预约和费用。此阶段按分支进入,不设全体等待路线确认的门槛。
- - 默认交付 `map.html`、`food.html`、`city-impression.html` 与 `index.html`;用户已明确只需部分成果时按其范围,不重复问版本。三页独立可打开,美食与城市名片不依赖地图 SDK 或密钥才能阅读。
- - `map.html` 使用高德 JSAPI 展示已确认路线,保留用餐时段,暂不加入美食攻略的具体餐厅点位;用户选中的美食街、夜市仍是路线景点。已有餐厅预约或明确必吃地点按实际约束保留,不能擅自删除。
- - `food.html` 展示区域/类型分组的菜品与店家卡片、选择理由、实用信息、来源与匹配图片,不强行按天分配;`city-impression.html` 展示城市特色与入选景点图文,不依赖每日游览顺序。缺图给清晰占位或文字卡,不伪造图片。
- - 共用当前数据与稳定地点编号,不在做页面时重排路线或重复研究。输出放在本行程目录,默认纯静态、不新增后端;在线地图依赖网络和有效配置。临时预览服务需说明用途及关闭方式,实际运行限制如实告知。
- - `assets/amap-html-template/` 提供不含具体行程和凭据的地图骨架。仅在本次行程目录使用它,按已确认数据填入标题、日期、点位、路段和说明;空骨架不算完成的地图。美食、城市名片与入口依据本次资料分别生成。
- - 地图默认 `pitch: 0`、`rotation: 0`,正面朝北。按天分色、显示顺序和方向箭头;合适缩放级别显示地名(旧版约 1.5 公里比例尺仅作体验参考),避免遮挡。住宿、交通节点与景点有区别;列表可定位,手机可读。
- - 点位详情展示当天时间、停留、开放/预约、交通和备注。区分点位顺序示意与真实道路路线;真实酒店替代住宿参考点,但附近街区仍保留为独立地点。
- - `index.html` 的导航与样式骨架可和子页同时制作;各页分别验收后接入,完成切换、独立打开/返回与手机检查再交付入口。入口只负责组织页面,不复制维护一份日程;分支尚未完成时准确说明状态,不把空页或失败的地图称为完整手册。
- **验收:** 浏览器检查地图非空、位置和顺序合理、默认朝向、方向箭头、缩放标签、日期筛选及列表定位;抽查 3—5 个点位与最新日程一致,检查手机遮挡。分别打开美食与城市名片检查内容、核心景点覆盖、图片对应与加载,再检查入口切换与公共信息一致。无法加载就报告对应页面未完成,不能仅以文件已生成作为成功。
- ## 继续与修改
- 读取已有数据、状态及本次相关资料 → 记录变化 → 补查受影响信息 → 重算相关日期/路段/费用 → 同步受影响文档与页面 → 汇报变化和待办。只换游览顺序不重做独立美食与城市介绍;变更入选景点、门店或事实时更新对应内容,维持共享信息一致。
- 只调整受影响部分,保留已发生记录和不受影响的决定;更换酒店同时检查每天首尾交通和费用,不能覆盖街区景点。已订安排若需要改变先说明代价,不把新草案当成已改签或已取消。出发前按需复核易变信息;未受托设置提醒时,不声称会自动监控或代办。
README.md · README · 279 lines
View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)
- https://github.com/user-attachments/assets/ff0209ed-4060-4cf2-8acd-9d37f2ab72ae
- <div align="center">
- # 🧭 Marco Polo
- **English** | [简体中文](README.zh-CN.md)
- ### Your next trip, mapped out.
- **An open-source AI travel assistant that turns your preferences into a trip you can actually follow.**
- Choose your stay and sights. Marco Polo connects real locations, transport, visiting hours, and reservations—then delivers an interactive map, food guide, and city guide in one travel handbook.
- 
- 
- 
- [](https://github.com/megumi-ben/marco-polo.skill/stargazers)
- **[Explore the live demo](https://megumi-ben.github.io/marco-polo.skill/)** · **[Get started](#quick-start)** · [Watch the video](https://www.bilibili.com/video/BV1qvaH66Em5/) · [Contribute](CONTRIBUTING.md)
- </div>
- [](https://megumi-ben.github.io/marco-polo.skill/nanjing/)
- <p align="center"><sub>Click to explore the Nanjing handbook · No setup needed · Demo content is in Chinese.<br>Screenshot: original Amap version. Public demo: OpenFreeMap. Lines show visit order; transport is queried separately during planning.</sub></p>
- <table>
- <tr>
- <td width="33%" align="center"><a href="https://megumi-ben.github.io/marco-polo.skill/nanjing/"><img src="assets/demo-handbook.jpg" alt="Nanjing travel handbook with tabs for routes, food, and the city guide" width="100%"></a></td>
- <td width="33%" align="center"><a href="https://megumi-ben.github.io/marco-polo.skill/nanjing/food.html"><img src="assets/demo-food.jpg" alt="Nanjing food guide with local dishes, restaurants, and category filters" width="100%"></a></td>
- <td width="33%" align="center"><a href="https://megumi-ben.github.io/marco-polo.skill/nanjing/city-impression.html"><img src="assets/demo-city.jpg" alt="Nanjing city guide with illustrated attraction cards and theme filters" width="100%"></a></td>
- </tr>
- <tr>
- <td align="center"><b>Travel handbook</b><br><sub>Routes, food, and sights in one place.</sub></td>
- <td align="center"><b>Food guide</b><br><sub>Find local dishes and places to eat.</sub></td>
- <td align="center"><b>City guide</b><br><sub>Get a feel for the city and every stop.</sub></td>
- </tr>
- </table>
- <p align="center"><sub>Click a preview to open the interactive example</sub></p>
- ## Why Marco Polo
- ### 📍 Plans grounded in real places
- **Routes start with coordinates.** Amap locations, entrances, and transport queries help group stops around your accommodation and reduce backtracking. Xiaohongshu adds travel experiences and local recommendations; FlyAI searches flights, trains, and hotels when needed.
- ### 🕒 Details that fit into a real day
- **Opening hours, bookings, meals, and travel time are planned together.** Food streets belong around meals, night views after dark. A reservation checklist tells you what to book and when, with time left for breaks and departure.
- ### 🗺️ A handbook you can use on the trip
- **See each day on an interactive map.** Explore place details, switch dates, and open the food and city guides from one entry point. Keep the itinerary, reservation checklist, and budget alongside it.
- ### 🎛️ Your trip, your choices
- **Compare first, choose next, refine together.** Get a varied shortlist with reasons and tradeoffs. Keep existing bookings, change your hotel, swap a place, or leave an afternoon free; the affected plan and costs are updated together.
- <details>
- <summary>Explore the details: routes, reservations, budgets, and revisions</summary>
- ### 📍 Routes grounded in real locations, planned around your stay
- Amap provides **POIs, coordinates, and entrances** for attractions, accommodation, and relevant transport hubs. The workflow checks ambiguous names and sites within larger attractions, then groups visits by location. Your chosen accommodation shapes where each day begins, which places belong together, and how you get back.
- Walking, public transport, or driving routes between consecutive stops supply road distances and estimated travel times. Detours, entrances, and transfers all inform the plan. **The geography and the journey both matter.**
- ### 🕒 Detailed days with time for sights, food, bookings, and breaks
- Museums during visiting hours, food streets around mealtimes, and evening views after dark: each stop is scheduled alongside opening hours, visit duration, reservation slots, and travel time. Meals, rest, queues, luggage collection, and getting to your train or flight also need room in the day.
- A separate **attraction reservation checklist** records booking channels, ticket release rules, target slots, current status, and alternatives if a booking falls through. Prepare ahead, then see the relevant reminders in each day's itinerary.
- ### 🗺️ A polished, interactive map you can explore
- The final **`map.html` webpage** shows each day in a different color, with visit order and direction arrows. Filter by date, locate a place from the itinerary, and open details about visit duration, reservations, and transport. Browser checks include readability on mobile.
- Accommodation, attractions, and transport hubs have distinct markers. Selected food streets and night markets stay on the route; restaurant recommendations live in a separate guide. See the whole trip before departure, then find your next stop while traveling.
- The **food guide** combines web research and Xiaohongshu experiences to recommend local dishes and specific restaurants, grouped by area or type. The **city guide** introduces the city and selected attractions with highlights, visiting tips, and matching images. Both pages can be read independently of the map.
- **Useful details throughout the process:**
- | Advantage | What it means for your trip |
- |---|---|
- | 🧭 **You keep the choices** | Candidates cover different interests and neighborhoods, with reasons, priorities, and tradeoffs. Explore options before narrowing them down. Existing tickets, hotels, and must-see places are respected. |
- | 🔎 **Sources and uncertainty stay visible** | Xiaohongshu contributes experience and opinions; official sources verify opening hours, tickets, and reservations. Sources and lookup dates are retained. Unverified facts and outstanding bookings stay clearly marked. |
- | 💰 **Clear costs per person** | Shared hotel and taxi costs are split across travelers; individual costs such as admission are recorded separately. Confirmed, estimated, and unknown amounts stay distinct, with checks against counting bundled tickets twice. |
- | 🔄 **Changes stay consistent** | Changing accommodation, places, or dates updates the affected itinerary, routes, reservation targets, costs, and map. Existing bookings retain their actual status, and changes that affect them are explained. |
- Your itinerary, reservation checklist, budget, map, and structured data are saved as local files. Keep the results and continue refining them.
- Use it for a weekend away, a trip with friends, or a few days of sightseeing after you've already booked transport and accommodation. Transport you are arranging yourself can be skipped; an existing hotel becomes the starting point for planning.
- </details>
- ## Quick start
- ### 1. Install Marco Polo
- 1. **[Download the Skill ZIP](https://megumi-ben.github.io/marco-polo.skill/downloads/marco-polo.zip)**.
- 2. Extract it and place the `marco-polo` folder in your project's `.agents/skills/` directory.
- The entry point should be `.agents/skills/marco-polo/SKILL.md`. The package contains only the skill instructions, agent metadata, output specification, and map template—no demo website, screenshots, deployment files, or Git history.
- ### 2. Connect travel research and maps
- Install **the Xiaohongshu skill bundle and both Amap skills**, then configure your own Amap credentials and Xiaohongshu login. Add **FlyAI** when you need transport or specific hotel searches. Place these skills alongside `marco-polo/`; see [Dependencies](#dependencies) for download links and setup.
- Your agent also needs web research access to check official travel information.
- ### 3. Describe the trip you have in mind
- For example:
- ```text
- $marco-polo I'd like two relaxed days in Beijing, with historic architecture and parks.
- I'll arrange travel to and from the city myself, but haven't chosen where to stay.
- Help me plan.
- ```
- Start with what you know. Marco Polo asks for the dates, group size, budget, and other information needed for the current stage. Later, you can continue with requests such as:
- > My hotel has changed to this address. Update the first and last journeys each day, along with their costs.
- <details>
- <summary>Other locations, updates, and output folders</summary>
- Use the **Skill ZIP** above for installation. GitHub’s **Code → Download ZIP** and `git clone` provide the full development repository, including the demo website; see [Contributing](CONTRIBUTING.md) if you want to work on the project.
- - For personal use across projects, you can install to `~/.agents/skills/marco-polo/`. See the [official Codex skill locations](https://learn.chatgpt.com/docs/build-skills). If the skill does not appear, start a new session or restart Codex.
- - To update, download a fresh Skill ZIP. Move your existing skill folder outside the skill directories as a backup, then extract the new package. Unzipping over the old folder would leave old demo files behind. Preserve any local configuration you have added.
- - Results are created progressively in `travel_plan/<city-trip-id>/` under your working directory. Revisions reuse the same data, so you can continue where you left off.
- </details>
- ## From one message to a trip
- *An illustrative conversation. Opening hours, tickets, and routes are verified for the actual travel dates.*
- **1. Describe the trip**
- > I'd like two relaxed days in Beijing, with historic architecture and parks. I'll handle travel to and from the city, but haven't chosen where to stay.
- Marco Polo asks for missing essentials such as dates and group size, compares 1–3 areas to stay, and offers a varied set of attractions with reasons, priorities, visit lengths, and initial booking information.
- **2. Choose your stay and places, then review the daily plan**
- > Let's stay around Wangfujing. I'd like the Palace Museum, Jingshan, Beihai, and the Temple of Heaven. Leave the others out for now.
- After checking opening hours and reservations, Amap locations, entrances, and transport help group nearby stops with time for meals, breaks, and departure. You review the proposed itinerary:
- | Day | Simplified example |
- |---|---|
- | Day 1 | Palace Museum in the morning → lunch and rest → Jingshan → Beihai |
- | Day 2 | Temple of Heaven in the morning → lunch → free time, with a departure buffer based on your train |
- **3. Confirm the route and take your trip files with you**
- Get **a daily itinerary, reservation checklist, per-person budget, and travel handbook**. The map, food guide, and city guide have separate pages, with `index.html` as the entry point. Map loading, navigation, images, and mobile layout are checked. Outstanding bookings retain their actual status, and you can continue revising the trip.
- <details>
- <summary>Deliverables and what they are for</summary>
- | Output | Purpose |
- |---|---|
- | Accommodation areas and attraction shortlist | Compare areas and places, with recommendation reasons and final selections |
- | Reservation checklist | Know when, where, and which time slot to book before departure |
- | Daily itinerary | Daily visits, transport, food, breaks, and alternatives |
- | Per-person budget (CSV) | Confirmed, estimated, and unknown expenses, calculated per person |
- | `map.html` | Locations, visit order, directions, and place details |
- | Food recommendations and `food.html` | Local dishes and specific restaurants, grouped by area or type |
- | `city-impression.html` | City character, selected attraction highlights, and images |
- | `index.html` | One entry point for the map, food guide, and city guide |
- | Trip data and progress | Facts, decisions, and progress for future revisions |
- Files are created as needed. Transport comparisons are added only when requested. See the [output specification](references/outputs.md) for the full file conventions.
- </details>
- ## Workflow
- Five stages turn preferences into a travel handbook. Research accommodation and attractions in parallel, then develop the route, food guide, and city guide independently. Only the map waits for route confirmation; content pages can be built as soon as their research is ready.
- ```mermaid
- flowchart TD
- A["1. Requirements and optional intercity transport"] --> B["2. Areas to stay"]
- A --> C["3. Attraction candidates"]
- B --> D["You choose accommodation and attractions"]
- C --> D
- D --> E["4. Route: verify opening and booking rules<br/>Amap locations, transport, timing, and costs"]
- D --> F["4. Food: web and Xiaohongshu research<br/>Dishes, restaurants, and practical details"]
- D --> G["4. City guide: city character<br/>Selected attraction highlights and images"]
- E --> H["You confirm the route"]
- H --> I["5. map.html"]
- F --> J["5. food.html"]
- G --> K["5. city-impression.html"]
- I --> L["index.html<br/>Check and deliver the complete handbook"]
- J --> L
- K --> L
- ```
- Existing transport and hotels are respected. Urgent reservation deadlines are surfaced early. Restaurant recommendations are independent of the daily route; selected food streets, night markets, and fixed reservations remain part of planning. Approving an itinerary does not place an order or make a reservation.
- ## Dependencies
- Install **both Amap skills and the Xiaohongshu skill bundle** for research and maps. Add **FlyAI** when you need transport, specific hotels, or travel products. Place dependencies alongside `marco-polo/`.
- | Skill | Purpose | Download |
- |---|---|---|
- | `amap-lbs-skill` | POIs, coordinates, entrances, and transport routes | [Official Amap guide](https://lbs.amap.com/api/skill/ready-to-use/summary) · [Official ZIP](https://a.amap.com/jsapi/static/openClaw/amap-lbs-skill.zip) |
- | `amap-jsapi-skill` | Interactive Amap webpages | [Official Amap guide](https://lbs.amap.com/api/skill/ready-to-use/summary) · [Official ZIP](https://a.amap.com/jsapi/static/openClaw/amap-jsapi-skill.zip) |
- | `xiaohongshu-skills` | Experience-based research on accommodation, attractions, and food | [XHS Bridge implementation](https://github.com/autoclaw-cc/xiaohongshu-skills); use Code → Download ZIP and preserve the repository structure |
- | `flyai` (optional) | Transport, specific hotels, and travel products | [Official setup](https://open.fly.ai/docs/quickstart) · [Source](https://github.com/alibaba-flyai/flyai-skill); use the `skills/flyai/` directory |
- <details>
- <summary>Amap setup and map preview</summary>
- **Amap:** Extract both ZIPs into your skill directory. Create your own Web Service Key and Web JSAPI Key in the [Amap console](https://console.amap.com/dev/key/app). Install the LBS runtime dependencies as documented (`npm install` for versions with a `package.json`) and configure `AMAP_WEBSERVICE_KEY`. JSAPI needs its own key and the corresponding security configuration. See the [Web Service setup guide](https://lbs.amap.com/api/webservice/create-project-and-key).
- When generating a map, copy `assets/amap-html-template/config.local.example.js` into the **trip output directory** as `config.local.js`. Set your own `jsapiKey` and either `securityJsCode` or an already deployed `serviceHost`. Keep the Web Service Key out of browser configuration. For public hosting, follow [Amap's security configuration guidance](https://lbs.amap.com/api/javascript-api-v2/guide/abc/prepare) for proxying and domain restrictions.
- To preview the handbook, run `python3 -m http.server 8000 --bind 127.0.0.1` in the trip directory and open `http://127.0.0.1:8000/index.html`; `map.html` also opens directly. The base map requires internet access and valid configuration. The food and city guides can be read independently. Press Ctrl+C in the server terminal when finished.
- </details>
- <details>
- <summary>Xiaohongshu installation and login</summary>
- **Xiaohongshu:** Requires Python 3.11+, uv, and Chrome. Run `uv sync` in the dependency directory. Follow the [upstream README](https://github.com/autoclaw-cc/xiaohongshu-skills#%E5%AE%89%E8%A3%85) to load its `extension/` directory as an unpacked Chrome extension and enable XHS Bridge, then run `uv run python scripts/cli.py check-login`. Authentication and search subskills are already included. Sign in with your own account; this travel workflow does not need publishing, comments, or likes.
- </details>
- <details>
- <summary>FlyAI setup (optional)</summary>
- **FlyAI (optional):** After installing `skills/flyai/`, run `npm i -g @fly-ai/flyai-cli`, then check `flyai --help` for the available commands. Follow the [upstream instructions](https://github.com/alibaba-flyai/flyai-skill#quick-start) for queries and optional API key setup, using your own credentials.
- </details>
- <details>
- <summary>Check that the dependencies work</summary>
- After installation, verify a POI and route query, plus a Xiaohongshu search and note-detail retrieval. A successful login alone does not verify the research capability. Missing capabilities are reported in the plan.
- </details>
- <details>
- <summary>Package structure and map template</summary>
- ```text
- marco-polo/
- ├── SKILL.md Five-stage workflow
- ├── README.md English overview and setup
- ├── README.zh-CN.md Chinese overview and setup
- ├── agents/openai.yaml Display name and invocation prompt
- ├── references/outputs.md Output conventions
- ├── docs/ Public website and interactive Nanjing demo
- ├── CONTRIBUTING.md Contribution and local preview guide
- └── assets/
- ├── demo-map.png Nanjing map showcase
- └── amap-html-template/
- ├── map.html Map template without itinerary data
- └── config.local.example.js Configuration placeholders
- ```
- </details>
- ## Help make travel planning better
- See the [contributing guide](CONTRIBUTING.md) for trip reports, focused pull requests, and local demo checks.
- Want to try it on your next trip? **[⭐ Star Marco Polo](https://github.com/megumi-ben/marco-polo.skill)** to find it again when you're ready to plan.
- [Share your experience](https://github.com/megumi-ben/marco-polo.skill/issues): what worked, where a route took a detour, or which booking detail was missing. Contributions to setup, route planning, and map presentation are welcome. Remove credentials, orders, and personal information before posting.
- Start contributing with the [planning rules](SKILL.md) and [output specification](references/outputs.md).
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.