- name
SKILL.md
Le fichier du skill reste dans le dépôt de l'éditeur.
Collez la commande copiée dans un terminal. Elle installe ce SKILL.md dans cet agent.
npx skills add nano-muse/nanoMuse --skill train-tickets -a cursor -yCollez la commande copiée dans un terminal. Elle installe ce SKILL.md dans cet agent.
npx skills add nano-muse/nanoMuse --skill train-tickets -a claude-code -yCollez la commande copiée dans un terminal. Elle installe ce SKILL.md dans cet agent.
npx skills add nano-muse/nanoMuse --skill train-tickets -a codex -yCollez la commande copiée dans un terminal. Elle installe ce SKILL.md dans cet agent.
npx skills add nano-muse/nanoMuse --skill train-tickets -a grok -yEnvoyez le message copié dans un chat Grok Bot. Le Bot récupère ce SKILL.md sur GitHub et l'ajoute aux skills que partagent tous vos Bots.
Install this skill: https://github.com/nano-muse/nanoMuse/tree/HEAD/nanomuse/skills/builtin/train-ticketsDe l'éditeur
Two hands, in this order. Searching is done with the 12306__* tools, a community MCP server (12306-mcp) that reads 12306's public timetable and seat counts; no login, no browser, no phone, nothing typed into an app. Booking has no API a personal agent may use; it happens on the phone's screen with phone_task, and only when the user has picked a train and said to book it.
Look for 12306__get-tickets in your tools (the tools list is the only place it can be: no shell command or script to check). If it is not there, say so in one line, give the user the lines to add to config.toml, and, when a phone is connected, offer the search on the phone's screen instead (the old way, below); otherwise stop:
[[mcp.servers]]
name = "12306"
command = "npx"
args = ["-y", "12306-mcp"]
risk = "safe"
egress = true
Node is needed for npx; the phone's own runtime has it.
recall (preferred stations, morning or evening, seat class, who travels) count as answers. Ask for the rest with one ask_user call.12306__get-current-date first whenever the user said 明天 / 后天 / 下周三; the server's date is the one that counts, not yours.12306__get-station-code-of-citys with citys = 北京|上海 gives the code that stands for a whole city (every station in it). A specific station (北京南, 上海虹桥): 12306__get-station-code-by-names. All the stations of a city: 12306__get-stations-code-in-city.12306__get-tickets with date, fromStation, toStation (names or codes), and the filters that fit: trainFilterFlags (G 高铁/城际, D 动车, GD both, Z/T/K the slow ones, F 复兴号), earliestStartTime / latestStartTime (hours, 0–24), sortFlag (startTime, arriveTime, duration), limitedNum (ask for 5–8, not the whole day). One call; a second only if the first came back empty (widen the hours or drop the flag).12306__get-interline-tickets with the same arguments, middleStation optional: the first ten connections. Where a train stops and when: 12306__get-train-route-stations with trainCode and departDate.Then, in your own words, the options that fit, three at most: 车次, 出发站 and time, 到达站 and time, 历时, the seat the user wants with its price and whether there are seats left (剩余 N 张 / 有 / 无). Say which you would take and why (earliest, shortest, cheapest that still has seats). Seat counts move by the minute; say so when the number is small.
Only when the user has picked a train and said to book it, and only with a phone connected (the system prompt says whether one is). Without a phone, say the train, date and price and that the booking is theirs to make in the app; offer a reminder for when sales open or a calendar entry.
phone_task, app = 铁路12306: open the chosen train on the chosen date, pick the seat class and the passengers the user named, and stop on the order confirmation screen; read back the train, date, passengers, seats, total. The goal must say: 停在提交订单页,不要点「提交订单」或任何支付按钮。phone_task for 提交订单; Sentinel will ask the user again for that tap; that is expected, not an error. Payment (支付) is the user's: say the order is placed and waiting to be paid in the app, and stop.Only when the server is not configured and a phone is connected. One phone_task, app = 铁路12306, with a goal like:
在 12306 首页把出发地设为「北京」、目的地设为「上海」、日期设为 9月25日,点「查询车票」。在结果页读出按发车时间最早的前 5 班:车次、发车站和发车时间、到达站和到达时间、历时、二等座价格和余票。只查询,不要点任何车次进入下单。
Rules for the operator you can put in context: the left station is 出发, the right one 到达, the swap icon sits between them; the date is picked on a calendar page; 筛选 / 只看高铁 narrow the list; the results list scrolls. Prices on the list page are the lowest class ("¥xxx 起"); say so.
amap skill) and the train make one plan: leave home at 07:10, G1 at 08:00.trips/<route>-<date>.md in the workspace.remember the stations and times the user keeps choosing, so the next search needs no question.Trains in China through the 12306 MCP server: routes, dates, options, connections and stops with no screen; booking only on the user's word, in the 12306 app on their phone, stopping before payment. U
---name: train-ticketsdescription: Trains in China through the 12306 MCP server: routes, dates, options, connections and stops with no screen…---# nanomuse/skills/builtin/train-tickets/SKILL.mdVue d'ensemble
Trains in China through the 12306 MCP server: routes, dates, options, connections and stops with no screen; booking only on the user's word, in the 12306 app on their phone, stopping before payment. Use when the user asks about 火车票 / 高铁 / 12306 or a train between two cities.
FAQ
Téléchargez nanomuse/skills/builtin/train-tickets/SKILL.md auprès de l'éditeur, ou copiez le chemin et ouvrez le dépôt. PlugMCP répertorie le skill et n'héberge pas le fichier.
L'éditeur le maintient sur https://github.com/nano-muse/nanoMuse.
Claude, Cursor, Codex et Grok Build peuvent lire un SKILL.md que vous ajoutez depuis le dépôt de l'éditeur. Grok Bot l'installe lorsque vous envoyez dans un chat le lien du dossier GitHub du skill.