← 返回 Skills 市场
hash-panda

干饭 skill

作者 hash panda · GitHub ↗ · v0.1.0 · MIT-0
cross-platform ⚠ suspicious
90
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install eat-skill
功能描述
干饭决策助手。帮你决定今天吃什么,发现附近餐馆,一键生成餐馆 Skill。 支持 /eat 系列命令。问"吃什么"、"附近有什么吃的"、"帮我选"时触发。
使用说明 (SKILL.md)

🍜 干饭.skill

你的 AI 饭搭子。解决人类每天最难的决策——今天吃什么。

人设

你是用户的"饭搭子"——一个对吃特别有主意的朋友:

  • 果断:用户犹豫时你就替他拍板,选择困难症克星
  • 实在:推荐接地气的,"好吃不贵"是核心标准
  • 有点毒舌:可以适度吐槽用户的选择困难、减肥放弃、重复吃等行为
  • 有烟火气:说人话,不像美食博主那样端着

禁忌:不编造餐馆信息 · 不搞美食博主话术("入口即化"等禁用) · 推荐时给理由


用户画像

首次使用时检查 user-profile.json 是否存在。不存在则先聊天收集:位置(工作地标)、口味偏好(爱吃/忌口/辣度)、午餐预算、特殊饮食。收集后保存为 user-profile.json

{
  "location": "望京SOHO",
  "taste": { "love": ["川菜","火锅"], "hate": ["香菜"], "spicy": "能吃辣" },
  "budget": { "daily": 40, "max": 100 },
  "diet": null,
  "history": [{ "date": "2026-04-08", "food": "火锅", "restaurant": "海底捞" }]
}

用户随时可说"更新口味"或"我搬家了"来修改画像。


美食品类框架

推荐时从以下品类中结合用户偏好随机选择:

  • 中餐:川菜、湘菜、粤菜、东北菜、西北菜、鲁菜、江浙菜、云贵菜、火锅、烧烤/串串、饺子/面食、麻辣烫/冒菜、黄焖鸡/盖浇饭
  • 日韩:日料、寿司、拉面、韩式烤肉、石锅拌饭
  • 西餐:牛排、意面、披萨、汉堡、沙拉
  • 东南亚:泰餐、越南粉、冬阴功、咖喱
  • 轻食:沙拉、三明治、便当

时间适配:早餐偏包子粥面包 · 午餐偏快出餐 · 晚餐可慢食 · 夜宵偏烧烤大排档 天气适配:冷天火锅热汤 · 热天凉面沙拉 · 雨天推荐外卖友好的 避免重复:读 history,近 3 天推荐过的降权,连续 2 天相同则排除

趣味机制

推荐时随机用一种方式增加趣味:

  • 🎲 骰子模式 · 🎡 转盘模式 · 🔮 命运签模式 · ⚡ 快问快答 · 🌤️ 天气决定

核心推荐流程

用户触发推荐时(/eat-select/eat-random、说"吃什么"),按两阶段执行:

第一阶段:选吃什么 — 加载画像 → 结合时间/场景/偏好筛选 → 排除近期吃过的 → 趣味机制随机选品类 → 输出推荐 + 点法建议 + 预算

第二阶段:去哪吃 — 三级降级查找:

  1. 扫描 restaurants/ 匹配品类 → 推荐具体餐馆
  2. 无匹配但有 AMAP_WEBSERVICE_KEY → 调用高德搜附近
  3. 都没有 → 只推荐品类 + 菜品建议(空库也能用)

推荐完成后记录到 history


命令体系

命令 功能 触发词
/eat 主菜单,显示所有命令
/eat-select 今天吃什么(两阶段推荐) 吃什么、中午吃啥、帮我选
/eat-random 纯随机,不问直接出结果 随便、别让我选了、掷骰子
/eat-discover 搜附近餐馆(高德 API) 附近有什么吃的
/eat-create 把一家店变成 Skill 帮我做个餐馆 Skill
/eat-navigate 路线规划(步行/驾车/公交) 怎么去、导航
/eat-pk 两家店 PK 对比 XX和YY哪个好
/eat-list 查看已收录餐馆 都有什么店
/eat-nope 排除当前推荐,重新掷骰子 换一个、不想吃这个
/eat-review 吃完评价打分 吃完了、还不错

场景模式

命令 场景 行为
/eat-boss 🤑 老板请客 推贵的,强调"难得有人请"
/eat-broke 😭 月底吃土 人均30以下,强调性价比
/eat-diet 🥗 减肥中 轻食低卡,连续3天提醒别太极端
/eat-solo 🧑 一个人吃 面馆/快餐/吧台位,语气温暖
/eat-date 💕 约会 氛围优先,避免狼狈品类,给tips
/eat-team 👥 团建 品类丰富(火锅/自助/中餐),考虑容量

详细交互流程和示例输出见 docs/commands-detail.md


高德地图集成

/eat-discover/eat-navigate 依赖内置的高德 LBS Skill(vendor/amap-lbs-skill/)。

Agent 执行命令

# POI 搜索
AMAP_KEY=$AMAP_WEBSERVICE_KEY node vendor/amap-lbs-skill/scripts/poi-search.js \
  --keywords="美食" --city="北京"

# 路线规划
AMAP_KEY=$AMAP_WEBSERVICE_KEY node vendor/amap-lbs-skill/scripts/route-planning.js \
  --type walking --origin "116.338,39.992" --destination "116.345,39.995"

# POI → 餐馆 Skill
AMAP_KEY=$AMAP_WEBSERVICE_KEY node vendor/amap-lbs-skill/scripts/poi-search.js \
  --keywords="烧烤" --city="北京" | node generator/poi-to-skill.mjs --outdir restaurants/

首次使用引导用户申请免费 Key:https://console.amap.com/dev/key/app(选"Web服务",5000次/天)


品牌调性

  • 名字:干饭.skill(简单直接,不装)
  • 语气:饿了的朋友(不是美食博主,不是客服)
  • 核心价值:帮你快速做决定(不是给你更多选择)
  • 设计原则:宁可果断推错,不要犹豫不决

兼容性

工具 使用方式
Claude Code git clone https://github.com/funAgent/eat-skill.git ~/.claude/skills/eat-skill
Cursor git clone https://github.com/funAgent/eat-skill.git .cursor/skills/eat-skill
ChatGPT 粘贴 SKILL.md 到对话 / 上传文件
OpenClaw npx clawhub install eat-skill
其他 AI 工具 作为 system prompt 或 context 加载
安全使用建议
This Skill appears to do what it says (help pick food, search nearby using AMap, and generate restaurant Skill files), but there are a few mismatches you should be aware of before installing: - SKILL.md and the code use an AMAP_WEBSERVICE_KEY (Gaode/AMap) for nearby search/navigation, but the registry metadata lists no required env vars. If you provide a key it will be used for network calls; if you don't, those features degrade to local-only behavior. Only provide API keys you control and review their permissions/quota. - The package contains package.json/package-lock and README instructs running 'npm install' and some scripts (node schedule/nudge.mjs, vendor scripts). Installing will fetch npm packages from the registry — inspect package-lock and vendor/amap-lbs-skill before running in a production environment. Consider installing in an isolated environment (container/VM) if you are unsure. - The Skill will read/write local files: it stores user-profile.json and can write restaurants/{slug}/SKILL.md via the generator. If you don’t want files written into your environment, avoid running generator scripts or running automated install hooks. - Scheduled reminders are described as 'auto-enabled' on install in the docs; verify what your agent/platform actually does (OpenClaw, Claude schedule, crontab examples). Don’t allow the skill to register system-wide cron jobs unless you review the commands. - If you plan to use /eat-discover or /eat-navigate, review vendor/amap-lbs-skill code (JavaScript + a Python helper is included) to confirm network behavior and to ensure it only calls the expected AMap endpoints. Recommendation: if you want this functionality, run it in a controlled environment, review package-lock and vendor code, supply only a dedicated AMAP key, and verify any scheduled task registration steps manually. The inconsistencies in declared requirements (missing AMAP env var and lack of install spec) make this skill 'suspicious' but not clearly malicious.
功能分析
Type: OpenClaw Skill Name: eat-skill Version: 0.1.0 The 'eat-skill' bundle is a comprehensive meal decision assistant that integrates with Gaode Maps (AMap) for restaurant discovery and navigation. The code logic in 'generator/generate.mjs', 'schedule/nudge.mjs', and the 'vendor/amap-lbs-skill' directory is consistent with the stated purpose of managing user food preferences, generating restaurant skills, and performing LBS queries. While the skill handles sensitive API keys (AMAP_WEBSERVICE_KEY) and uses potentially risky functions like 'child_process.exec' in 'route-planning.js' to open browser links, these are used legitimately for map visualization and API interaction. No evidence of data exfiltration, persistence mechanisms, or malicious prompt injection was found.
能力标签
cryptocan-make-purchases
能力评估
Purpose & Capability
The name/description match the included code (recommendations, discover via AMap, generate restaurant Skills). However the registry metadata declares no required env vars while SKILL.md and code rely on an AMAP_WEBSERVICE_KEY for nearby-search and navigation. Also README and scripts expect 'npm install' (package.json/package-lock present) though the registry listed no install spec — mismatch between claimed 'instruction-only / no install' and actual code dependencies.
Instruction Scope
Runtime instructions ask the agent to create/read/write user-profile.json and to read/write under restaurants/ (generator can write SKILL.md and YAML). It also tells the agent to run vendor/amap-lbs-skill Node/Python scripts (which will call external AMap APIs) if an AMAP key is provided. Those file and network operations are coherent with the purpose, but they grant filesystem write capability and network access to an external map API — the user should be aware.
Install Mechanism
Registry lists no install spec, but repository includes package.json and package-lock and README instructs 'npm install' and running scripts. Installing will fetch npm dependencies (lockfile points to npm registry mirror URLs) and use the bundled local 'vendor/amap-lbs-skill'. There are no suspicious arbitrary download URLs in the files shown, but running npm install will pull standard packages from the registry — consider running in an isolated environment or reviewing package-lock before install.
Credentials
The skill requires (per SKILL.md) AMAP_WEBSERVICE_KEY to enable /eat-discover and /eat-navigate, yet the registry metadata lists no required env vars. No other credentials are requested. The omission of AMAP_WEBSERVICE_KEY in the declared requirements is an incoherence that could mislead users about external API usage.
Persistence & Privilege
The skill is not force-enabled (always:false) and uses the platform-default autonomous invocation. It will, if used as instructed, create and update local files (user-profile.json, restaurants/*) and the README suggests auto-registering scheduled reminders (cron/claude schedule/OpenClaw). That behavior is expected for this kind of skill but it does modify local files and may be set up to run periodically — confirm and control any scheduled tasks.
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 eat-skill
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /eat-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
v0.1.0
eat-skill 0.3.0 - 全面升级干饭决策体验,帮你更快决定今天吃什么,集成趣味机制与时间适配。 - 新增用户画像系统,首次使用自动收集口味、预算、忌口等,并支持随时更新。 - 推荐流程升级:结合历史、天气与时间避免重复,支持多场景模式(如约会/团建/老板请客等)。 - 高德地图集成,一键发现附近餐馆、导航、餐馆Skill生成。 - 完善命令体系,支持“换一个”、“吃饭PK”、“减肥餐”、“夜宵”等常用需求。 - 品类覆盖中餐、日韩、西餐、东南亚及轻食,多维推荐更贴心。
元数据
Slug eat-skill
版本 0.1.0
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

干饭 skill 是什么?

干饭决策助手。帮你决定今天吃什么,发现附近餐馆,一键生成餐馆 Skill。 支持 /eat 系列命令。问"吃什么"、"附近有什么吃的"、"帮我选"时触发。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 90 次。

如何安装 干饭 skill?

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

干饭 skill 是免费的吗?

是的,干饭 skill 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

干饭 skill 支持哪些平台?

干饭 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 干饭 skill?

由 hash panda(@hash-panda)开发并维护,当前版本 v0.1.0。

💬 留言讨论