跳到正文
cccskills
免费在 GitHub 公开

places_search

Find and compare restaurants, hotels, parks, shops, and local services in a specified area. Check details such as opening hours and photos, then present relevant options with a map when the required location data is available.

中文简介正在准备中。当前显示英语简介。

原文查看原文说明

Find, compare, and share details on physical places near the user or in a specified area, including restaurants, cafes, bars, hotels, parks, attractions, shops, and businesses with local services. Not for itineraries, choosing a city or region, dated events or showtimes, or directions.

实测评分 未实测

查看安装方法

适用场景

  • Find restaurants that match an area and food preference
  • Compare hotels or cafes in a neighborhood
  • Check a venue's opening hours and photos
  • Identify a place using captured images and GPS

包含的文件(2)

  • SKILL.md6.7 KB
  • eval/scenarios.yaml4.2 KB

SKILL.md(原文)

安装前可以查看交给 AI 代理的指令内容。

Places

Use browser.search for all place discovery and name/address lookup. Run places <command> through muse.exec for richer details and photo-based place detection. Never load or search for a places tool namespace. If a search comes back empty or weak, refine the browser.search query per the rules below. Print CLI JSON to stdout for you to read, not to show the user. Use --help when needed. No sign-in is required.

Common flows

Find places

Use browser.search for standing places people can visit or reference on a map: restaurants, cafes, bars, hotels, parks, trails, attractions, shops, and local services. A follow-up like "anything cheaper?" or "Italian instead?" refines the preceding search.

  • Put the complete user intent in primary_query.query. Include the place category or venue name, area, radius, and ranking preference in natural language. For example, search for "best sushi restaurants in San Francisco within five miles", not just "sushi".
  • Put "near me" or "nearby" in a browser.search query only when message_location or last_seen_location arrived with this turn; the search service cannot see a location you learned any other way, so those words can land results in the wrong city. When the location came from the user, a device read, or their home on file, put the area's name in the query. If no location is available, ask the user; never invent or guess a location.
  • One call covers one area. When the user names multiple cities or neighborhoods, run a separate call for each and cover each in the answer.
  • For category discovery, include the category and area in the query. For a named venue, include its exact name and enough area context to disambiguate it.
  • For "best," trending, insider, or nuanced requests, use the web evidence and place results returned together by browser.search. Ground every named venue in a returned result.
  • When comparable avg_rating values from places details are available for every shortlisted place, order them highest first but never print the rating. Otherwise order by relevance to the user's constraints.
  • Avoid showing permanently or temporarily closed places unless specifically asked for.
  • Judge the returned result set rather than following rank blindly. Set aside wrong-category or wrong-area results. If the set is weak, retry with an exact venue name, a narrower subtype, or clearer area language.

Get richer details

  • places details takes numeric place IDs only. It cannot search by name or address; use browser.search for that.
  • Call places details only when browser.search explicitly returns an all-digit place_id for that result. Never derive an ID from a venue name or URL.
  • Prefer the browser.search response when it already contains enough detail. Use places details for richer or fresher hours, prices, photos, reviews, and offerings.
  • If browser.search returns no numeric ID, retry once with the exact venue name and area. If that still returns no ID, answer from cited search evidence without details or a map.
  • Batch several IDs into one call when comparing places. Pass --motivation with the user's intent so quotes and photos are ranked for it.
  • places details returns {"<place_id>": {"details": {…, "rating": {"avg_rating", "num_rating"}}}}, keyed by id.

Show places on a map

Before creating a map, run places details for the selected numeric IDs. Create one map if at least one returned details record has both a numeric place ID and coordinates. This applies to recommendations, comparisons, named-place lookups, and other uses where a map could ground the user. See Response formatting for which places go on it.

Create one local_map widget with widget.create and this payload:

{
  "kind": "local_map",
  "data": {
    "elements": [
      {
        "kind": "rich_place",
        "place_id": "<numeric ID>",
        "name": "<name from the same result>",
        "coordinate": {
          "latitude": 0.0,
          "longitude": 0.0
        }
      }
    ]
  }
}
  • Replace the example coordinates with the values from the same keyed places details record. Use its canonical name. Omit a place whose details or coordinates are missing; never infer an ID or geocode a name.
  • If creation fails, answer in text. Do not retry or build an HTML or image map.

Identify a place from a photo

Use places detect when captured frames and GPS need to resolve which place the user is at. It returns ranked candidates as {place_id, place_name, confidence}. Pass a returned ID to places details for more. The optional 4th --location field is the connected Wi-Fi BSSID; include it verbatim when known to improve Home/Work matching, and never echo a raw BSSID back to the user.

Response formatting

  • Never name a place that did not come back from a tool.
  • Match the framing to the ask. When the user wants a recommendation, give them two to three. When the request is broad, offer a few named results as a starting point.
  • Name two to three places in text from the merged result set. Write each on its own line in exactly this shape: **Name**, a sentence or two on why you chose it. Open with one short line framing the answer before the places.
  • Send the text as one message. Never add a second message after the map.
  • No addresses, raw place IDs, coordinates, star ratings, review counts, or price levels in text.
  • After a recommendation, note there are more options without itemizing; after a broad answer, offer to narrow to a recommendation.
  • Cite factual claims sourced from browser.search using its normal citation format.
  • Resolve every name you plan to use, closing line included, through browser.search before creating the map. Never name an unresolved place.
  • Build the map from the most relevant places returned that have a place ID and coordinates. Don't rely solely on named places in your text as an indication of what places should appear on the map.
  • The places you named come first on the map, in the order you named them, ahead of everything else.
  • Create one map per answer. Include the returned embed_token.
  • Never narrate or reference the map. No "I mapped things out for you"; "Take a look at the map", etc.

Limits

  • Do not write directions, turn-by-turn steps, maps links, or navigation controls.
  • Do not estimate travel times or distances. Use only values returned by a tool.

评价

还没有评价。欢迎分享你的使用感受。

评论

无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。

还没有评论。来写第一条吧。

提交后会立即公开。请勿填写邮箱、电话等个人信息。URL 最多 2 个。

0 / 1000 字

此表单受 reCAPTCHA 保护,适用 Google 的隐私权政策和服务条款。

同一仓库的技能

3d-object

免费有简介

使用three.js创建立体模型,在交互式查看器中从各个角度检查形状和材质,并以OBJ + MTL或GLB格式保存,供其他三维制作软件使用。

  • 创建立体模型并从各个角度检查
  • 制作可在Blender中继续编辑的模型
  • 检查模型轮廓和材质效果
asgeirtj/system_prompts_leaks7万2026年10月23日 更新

action-items

免费有简介

在个人任务清单中记录待办事项、承诺、期限、下一步、待回复对象和阻碍原因,更新进度与依赖关系,合并重复项并归档已结束的任务。

  • 记录有期限的任务与承诺
  • 整理待回复事项及阻碍原因
  • 前置任务完成后更新依赖关系
asgeirtj/system_prompts_leaks7万2026年10月23日 更新

agents

免费

Coordinate a task from a verifiable goal through completion. Handle work directly or split independent tasks across separate agent sessions, track progress, verify results, and arrange pull request merging when included in the goal.

中文简介正在准备中。当前显示英语简介。

  • Split independent tasks across agents
  • Check progress across ongoing tasks
  • Resume a previously started project
asgeirtj/system_prompts_leaks7万2026年10月23日 更新

animated-video

免费有简介

将动画制作成可播放、暂停和定位时间的HTML页面,通过时间轴调整场景时长与速度,并为横屏或竖屏视频导出做好准备。

  • 制作产品操作演示动画
  • 用文字和图形解释内容
  • 制作竖屏动画
asgeirtj/system_prompts_leaks7万2026年10月23日 更新

apple_healthkit

免费有简介

按指定时段查看已同步到Apple Health的步数、心率、睡眠和锻炼记录,了解日常变化,并检查同步状态及记录缺失情况。

  • 查看步数与心率变化
  • 检查睡眠阶段、时长及夜间醒来情况
  • 回顾运动时长、距离与消耗热量
asgeirtj/system_prompts_leaks7万2026年10月23日 更新

artifact_document

免费有简介

创建和编辑Word文档及模板,提取或整理内容、替换图片,并处理修订和批注。使用规范的标题与列表结构,交付前检查所有页面的显示效果。

  • 创建结构规范的Word文档
  • 整理现有文档的内容
  • 替换文档中的图片和文字
asgeirtj/system_prompts_leaks7万2026年10月23日 更新

查看 asgeirtj 的全部技能

报告此技能的问题