旅游攻略规划技能

飞猪旅游攻略规划技能（性能优化版 v2.0）。使用场景：(1) 从小红书搜索目的地旅游攻略，(2) 自动提取攻略中的目的地、日期、行程信息，(3) 查询飞猪机票价格并插入攻略中，(4) 生成完整的行程规划报告（含攻略 + 最低价机票链接 + 天气预报 + 详细预算）。**性能目标：3-4 分钟完成**（优化前 9...

Security Scan

The skill's behavior (browser scraping, Fliggy login/session use, and cron-based price monitoring) is coherent with a travel planner, but it fails to declare needed credentials, configuration, and persistence requirements and asks the agent to access browser sessions and external sites without making those requirements explicit.

Purpose & Capability

The name/description (scrape Xiaohongshu, query Fliggy, add weather, produce reports) matches the code and instructions. However the skill references needing a '飞猪账号配置' and uses a browser profile/session for logged-in actions (QR login) — yet declares no required env vars, primary credential, or config paths. That omission is disproportionate: querying Fliggy for booking URLs and maintaining logged-in sessions normally requires credentials or persistent session storage.

Instruction Scope

SKILL.md and references instruct the agent to open external sites (xiaohongshu, fliggy), snapshot/parse DOM, run curl to wttr.in/Open-Meteo, and extract note contents and booking links. They also require maintaining browser login (QR code) and choosing real note links. These runtime instructions implicitly require access to a browser tool and persistent sessions/cookies; those capabilities and the exact data flows (what is stored, cached, or sent elsewhere) are not declared. The instructions also include automated cron-style price monitoring and potential notifications, which imply background persistence or scheduling that is not described in installation/configuration.

Install Mechanism

No install spec (instruction-only) and included Python scripts are readable and not obfuscated. This minimizes installer risk — nothing is downloaded from arbitrary URLs. The scripts perform HTTP(S) calls and promise-placeholders for browser tool use; there is no malicious install behavior in the provided files.

Credentials

The skill requests no environment variables or credentials, yet the README/SKILL.md explicitly say a Fliggy account must be configured and that the built-in browser profile will be used (including QR login). This is a mismatch: credentials/session persistence required by the task are not declared. The skill also mentions caching and cron monitoring (persisted data), but doesn't declare storage/config paths or how secrets/sessions are stored/protected.

Persistence & Privilege

always:false (good). But the skill expects to keep browser sessions, cache weather/train data, and supports cron-style price monitoring (examples in references). Those behaviors imply persistence and possibly background scheduled tasks; the SKILL.md doesn't explain how scheduling/notifications are authorized or where caches/sessions live. That gap increases the security surface if the agent/platform persists session cookies or stored credentials for this skill without user visibility.

What to consider before installing

This skill appears to do what it says (scrape Xiaohongshu, query Fliggy, add weather, build reports), but there are gaps you should clarify before installing or using it: - Credentials and sessions: The docs say you need a Fliggy account and that the browser profile will be used (QR login). However the skill does not declare any required environment variables or config paths for storing credentials or cookies. Ask the maintainer: how are Fliggy credentials or session cookies provided and stored, and where is that data kept? If you must supply tokens or allow a persistent browser profile, understand how to revoke them. - Browser/tool access: The skill expects an agent-hosted browser tool capable of opening pages, snapshots, and maintaining login sessions. Confirm your environment provides such a tool and that you consent to the skill using it to access external sites and capture DOM content. - Persistence/cron: The references include price-monitoring cron examples and caching. Verify whether the platform will run scheduled tasks on your behalf and where cached data (including any booking URLs or saved pages) is stored and who can access it. - Privacy & ToS: The skill scrapes user-generated content (小红书) and automates interactions on Fliggy — make sure you are comfortable with scraping and that these activities comply with the target sites' terms of service. Be prepared for captchas and manual QR login steps. - Safety steps before use: request clear docs (TOOLS.md or secure config instructions) showing how to supply/revoke Fliggy credentials or session cookies; test the skill in a sandbox account if possible; and avoid granting persistent secrets until you confirm how they are protected. If you want limited risk, consider forbidding autonomous invocation for this skill or reviewing/logging browser session activity.

飞猪旅游攻略规划技能 v2.0（性能优化版）

自动从小红书获取旅游攻略，并整合飞猪机票信息，生成完整的行程规划！

⚡ 性能优化： 真正并行查询 + 数据缓存 + 精简提取，耗时从 9 分钟降至 3-4 分钟！

---

🔗 技能依赖

1. weather skill（天气查询）

依赖原因： 获取目的地天气预报

使用场景：

• 出行日期在未来 3 天内 → 使用 weather skill 查询真实天气
• 出行日期在未来 3 天后 → 使用气候数据估算，并标注"基于气候数据估算"

调用方式：

curl -s "wttr.in/{城市名}?format=j1"

2. flight-price-comparison skill（机票查询）

依赖原因： 查询飞猪机票价格

---

⚡ 性能优化 v2.0

优化后执行流程（目标：3-4 分钟）

用户输入："4.3 杭州到大理 4 天 3 晚攻略"
     ↓
t=0s    真正并行发起 3 个查询
        ├─ browser.open("小红书", 60 秒超时)
        ├─ exec("curl wttr.in/大理", 10 秒超时)
        └─ browser.open("飞猪", 60 秒超时)
     ↓
t=10s   天气查询完成 ✅
     ↓
t=60s   页面打开完成，并行 snapshot（30 秒）
     ↓
t=90s   数据抓取完成
     ↓
t=120s  精简数据提取（只取 5 个核心字段，30 秒）
     ↓
t=180s  生成行程文档（60 秒）
     ↓
✅ 完成！总计约 3 分钟

核心优化点

优化点	优化前	优化后	节省
查询方式	串行执行	真正并行	2-3 分钟
Snapshot	完整 DOM（3000+ 元素）	精准提取关键元素	1 分钟
天气查询	每次 API	优先缓存	10-30 秒
数据提取	20+ 字段	5 个核心字段	20-30 秒
总计	~9 分钟	~3-4 分钟	60%+

并行查询实现（真正并行）

// ✅ 正确做法：真正并行发起
const [xiaohongshuTab, weatherData, fliggyTab] = await Promise.all([
  browser.open({ 
    url: "https://www.xiaohongshu.com/search_result?keyword={目的地} 旅游攻略",
    timeoutMs: 60000
  }),
  exec({ 
    command: 'curl -s "wttr.in/{城市名}?format=j1"',
    timeout: 10
  }),
  browser.open({ 
    url: "https://sjipiao.fliggy.com/flight_search_result.htm?tripType=1&depCity={出发地}&arrCity={目的地}&depDate={日期}",
    timeoutMs: 60000
  })
])

// 并行抓取数据
const [xiaohongshuData, fliggyData] = await Promise.all([
  browser.snapshot({ targetId: xiaohongshuTab.targetId, refs: "aria", timeoutMs: 30000 }),
  browser.snapshot({ targetId: fliggyTab.targetId, refs: "aria", timeoutMs: 30000 })
])

数据缓存策略

可缓存的数据：

const CACHE = {
  // 气候数据（季节性，基本固定）
  weather: {
    "大理_春季": { temp: "8-22℃", weather: "晴到多云", tips: "紫外线强，带防晒" },
    "北京_春季": { temp: "10-20℃", weather: "多云", tips: "防风沙" },
    "三亚_春季": { temp: "22-30℃", weather: "晴朗", tips: "带泳衣" },
  },
  
  // 高铁价格（相对固定）
  trains: {
    "杭州 - 北京": { time: "4.5-6 小时", price: "¥550-650" },
    "杭州 - 大理": { time: "无直达，昆明中转", price: "¥550-650" },
  }
}

精简数据提取（只取核心字段）

优化后 Prompt（30 秒完成）：

只提取 5 个核心信息：
1. 行程天数（如"4 天 3 晚"）
2. 必去景点 Top3（数组）
3. 住宿推荐区域
4. 人均预算范围（不含机票）
5. 最佳季节

---

🚀 快速开始

使用示例

# 简单查询
帮我规划一下去大理的旅行，看下小红书的攻略和机票价格

# 指定日期
查一下 3 月 20 日去三亚的攻略，顺便看看机票多少钱

# 往返行程
计划 4 月 1 日 -4 月 5 日去日本旅游，帮我找攻略和机票

---

📋 核心流程

1️⃣ 小红书攻略搜索

• 搜索目的地相关攻略
• 提取高互动内容（点赞/收藏>平均值）
• 解析攻略关键信息

2️⃣ 信息结构化处理

使用大模型提取核心数据（5 个字段）

3️⃣ 飞猪机票查询

• 查询往返机票价格
• 提取最低价航班信息

4️⃣ 生成整合报告

输出完整行程规划

---

📝 行程输出标准格式

必须包含的内容（按顺序）

## ✅ 行程规划完成

### 🚄 交通推荐

| 方式 | 时间 | 价格 | 推荐 |
|------|------|------|------|
| [方式 1](链接) | X 小时 | ¥XXX | ⭐推荐 |
| [方式 2](链接) | X 小时 | ¥XXX | 备选 |

### 🗓️ 行程亮点

- **Day 1:** ...
- **Day 2:** ...
- ...

### 🌤️ 天气预报

| 日期 | 天气 | 温度 |
|------|------|------|
| X 月 X 日 | 晴/多云 | XX-XX℃ |

### 💰 总预算

| 类别 | 费用 | 说明 |
|------|------|------|
| 交通 | ¥XXX | 高铁/机票往返 |
| 酒店 | ¥XXX | X 晚住宿 |
| 门票 | ¥XXX | 景点门票 |
| 餐饮 | ¥XXX | X 天餐费 |
| 其他 | ¥XXX | 应急备用 |
| **总计** | **¥XXX** | 经济型 |

### 📖 小红书攻略来源

[查看小红书原文](链接)（XX 赞）

---

⚠️ 注意事项

1. 并行查询： 必须同时发起所有查询，不要串行
2. 超时控制： 所有网络请求设置超时（browser 60 秒，curl 10 秒）
3. 缓存优先： 天气/高铁价格优先查缓存
4. 精简提取： 只取 5 个核心字段，不要贪多
5. 小红书链接格式： 必须返回 /explore/{noteId} 格式的真实笔记链接，不要返回 /search_result/{id} 链接（会显示"安全限制"）

---

🔧 故障排查

Q1: 执行时间超过 5 分钟

• 检查是否真正并行执行（用 Promise.all）
• 检查 snapshot 是否设置了 timeoutMs
• 检查是否在使用完整 DOM 抓取

Q2: 数据提取不准确

• 增加参考笔记数量到前 5 篇
• 检查 Prompt 是否过于复杂

Q3: 小红书链接打不开（显示"安全限制"）

• 原因：返回了 /search_result/{id} 链接
• 解决：从 snapshot 中提取笔记卡片的真实链接（/explore/{noteId} 格式）
• 正确示例：https://www.xiaohongshu.com/explore/69b01f9500000000150236a9
• 错误示例：https://www.xiaohongshu.com/search_result/69b01f9500000000150236a9 ❌

---

版本： v2.0（性能优化版）
最后更新： 2026-03-16
维护者： Luna 🌙
性能目标： 3-4 分钟完成行程规划