← 返回 Skills 市场
tencent-adm

腾讯出行服务跑腿 Skill

作者 腾讯开源 · GitHub ↗ · v1.0.1 · MIT-0
cross-platform ⚠ suspicious
37
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install tencent-delivery-skill
功能描述
Books same-city on-demand courier delivery (pickup-and-deliver within 2 hours, typically small parcels <30kg) via Tencent Mobility MCP. Use this skill when t...
使用说明 (SKILL.md)

腾讯出行服务跑腿助手

帮你完成跑腿下单、查单、取消、查骑手等操作。


\x3Ca id="entry-hard-rule">\x3C/a>

0. 入口硬铁律

🛑 不论用户首句说什么,回复前先跑一条命令:

python3 ./scripts/tms_delivery.py bootstrap

按返回的 stage 分流(所有分支都已包含 next_doc / reply_template,直接照用):

stage 动作
ready §1 流程选择器
has_ongoing_order 原样输出 reply_template 拦截用户,禁下新单
setup_token quick-start-workflow 引导取 Token
error error-handling 处理

bootstrap 命令本身执行失败时command not found / 非零退出码且无 JSON stdout / 报 Python 相关错误),原样回复以下模版,禁自行揣测原因

当前环境无法执行跑腿 skill 所需的 Python 脚本,请按以下顺序自检:

1. 终端运行 `python3 --version`,确认已安装 Python 3.8+
2. 在当前 IDE / 客户端设置中确认已允许执行 shell 命令
3. 仍无法解决请查阅项目 README 的「环境前置」与「故障排查 FAQ」章节

1. FSM 驱动器

python3 ./scripts/tms_delivery.py state next   # 查下一步
python3 ./scripts/tms_delivery.py state show   # 查当前快照(手机号脱敏)

⚠️ 正常路径不必调 state next——commit-* / pick-address / select-sku / run-estimate 等命令返回值里已内含 next_state / next_action / next_doc(LLM 下一步该读的文档锚点)。只在异常分叉(用户中途"算了")时才调。


\x3Ca id="flow-selector">\x3C/a>

2. 流程选择器

用户意图 执行流程
跑腿下单 delivery-workflow.md
查订单 / 查骑手 order-workflow.md#order-query-flow
取消订单 order-workflow.md#order-cancel-flow
放弃当前流程 delivery-workflow.md#delivery-exit-flow
配置 / 换 Token quick-start-workflow.md

3. 核心约束

  1. 工具即权威:地址/价格/订单必须来自工具返回
  2. 不跳步不并行:串行执行,WRITE_GUARD 硬拦截
  3. 首选高阶命令bootstrap / pick-address \x3Crole> \x3CN> / select-sku \x3CN>——每条都内置多步逻辑,勿拆成 state set 序列
  4. 状态走 session.json:不从 chat history 捞字段
  5. 模版即全部:返回值含 reply_template 时原样粘给用户,禁自行改写
  6. 第一步只搜寄件、第二步只搜收件:禁预调用
  7. 输入侧不捏造:工具入参(地址关键词 / 手机号 / 联系人 / 物品类目 / 序号)必须来自以下三种合法来源之一:① 用户当轮消息原文里的词;② 脚本返回值 / session.json;③ assets/PREFERENCE.md 地址簿里的别名或联系人名(用户说"公司/家/妈妈家/张三"时直接把原词传给 resolve-address,脚本会自动查地址簿)。严禁基于背景知识、常识、IP 归属、跨会话历史推断(例:用户没说"腾讯总部"就不能拿"北京腾讯总部"去搜)。三种来源都不命中 → 先开口问,宁问勿造。
  8. 联系人字段允许乱序预登记:用户首条消息若带任何联系人字段(姓名 / 手机号,寄/收均可),第一时间调 prefill-contacts 一次性落盘 4 个槽位(未抽到的传空串)。后续 FSM 推进时已填字段自动跳过,可省去单独问联系人的轮次。详见 step-1-sender §0.1

\x3Ca id="output-leak-firewall">\x3C/a>

4. 防泄露铁律

🚫 用户端回复禁止出现estimatePriceRecordId / skuMap / skuId / selectedSkuId / 形如 60796_2{数字}_{数字} 值 / expressSkuInfos / defaultChecked

脚本出口已黑名单过滤,但 LLM 仍须自检:回复前扫描上述字段 → 发现即删。


5. 安全约定

  • Token 仅通过 save-token 写入、mcp-call 内部使用,禁输出到对话
  • Session 仅通过 state 子命令读写,禁 read_file / cat ~/.config/tms-delivery/session.json

6. 长尾查阅索引

脚本返回值里的 next_doc 就是 LLM 下一步该读的文档锚点,按提示读即可。无 next_doc 时按下表按意图查:

遇到什么情况 读哪个文档
用户表达跑腿 / 下单意图 references/delivery-workflow.md
查订单 / 查骑手 / 取消 references/order-workflow.md
配置 Token / 切换账号 references/quick-start-workflow.md
看到未识别错误码 / message references/error-handling.md
MCP 工具入参 / 响应字段细节 references/api-contract.md
调试 session 字段 / 状态机 references/session-state.md
安全使用建议
Install only if you are comfortable giving this skill a Tencent delivery token, letting it store that token and delivery-session details on disk, and allowing it to book, query, and cancel courier orders through Tencent MCP. Review the generic MCP call surface and the third-party payment QR generation before using it with real payment links or sensitive addresses.
能力评估
Purpose & Capability
The stated purpose of same-city courier booking, tracking, payment, and cancellation matches most artifacts, but the script includes a generic `mcp-call <tool>` path that forwards caller-supplied tool names with the user's token, which is broader than the delivery workflow needs.
Instruction Scope
The workflow requires running a local bootstrap command before replying and repeatedly tells the agent to paste templates verbatim, which is coherent for state management but can suppress user-facing privacy or payment warnings; some confirmation triggers such as `OK` are also broad for payment/cancellation-adjacent steps.
Install Mechanism
The package consists of markdown instructions and one Python helper script with no dependency install or obfuscated code; static scan and VirusTotal telemetry were clean. The artifacts do not provide a clear permission manifest for shell execution, local storage, network use, and credential handling.
Credentials
Tencent MCP network access, local session files, and address lookup are proportionate to courier booking, but payment QR rendering sends externally supplied payment URL data to a third-party QR generation service without a visible user warning or domain validation in the artifact.
Persistence & Privilege
The skill saves a reusable token in `~/.config/tms-delivery/env.json` and delivery session data in `~/.config/tms-delivery/session.json`; address-book saving is user-triggered, and deletion commands exist, but storage scope, sensitivity, and lifetime are under-explained to users.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install tencent-delivery-skill
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /tencent-delivery-skill 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.0.1
- Clarified skill scope to strictly same-city, ≥2-hour courier tasks (“跑腿”) within China, excluding cross-city shipping, food delivery, ride-hailing, and finance queries. - Now requires running a bootstrap command before every user interaction to determine state and response path; all reply templates and branches must follow tool output strictly. - Enforced tool-first data sourcing, input authenticity (no assumptions or invented info), and output sanitization (filtering restricted fields). - Improved error handling instructions for script execution failures, with specific troubleshooting steps. - Expanded FSM and workflow details, supporting flexible user input order and robust state/session management. - Added strong privacy and output leakage barriers to prevent exposure of sensitive identifiers or technical metadata.
元数据
Slug tencent-delivery-skill
版本 1.0.1
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 1
常见问题

腾讯出行服务跑腿 Skill 是什么?

Books same-city on-demand courier delivery (pickup-and-deliver within 2 hours, typically small parcels <30kg) via Tencent Mobility MCP. Use this skill when t... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 37 次。

如何安装 腾讯出行服务跑腿 Skill?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install tencent-delivery-skill」即可一键安装,无需额外配置。

腾讯出行服务跑腿 Skill 是免费的吗?

是的,腾讯出行服务跑腿 Skill 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

腾讯出行服务跑腿 Skill 支持哪些平台?

腾讯出行服务跑腿 Skill 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 腾讯出行服务跑腿 Skill?

由 腾讯开源(@tencent-adm)开发并维护,当前版本 v1.0.1。

💬 留言讨论