练习开发Skill——网页内容抓取Skill（website-content-fetch）

现在使用AI帮我们找一些资料帮我们分析问题的场景多的数不胜数，但是在AI找资料的过程中，我们对AI抓取的内容是不知道，也不可以明确指定范围的，主要是靠模型本身能力去收集，当然也可以增加提示词，加以控制。

当然目前解决方案也有很多：

• 增加更详细的提示词，描述更细致，控制更精细，过程更明确
• 同时也有Tavily Search、SearXNG等搜索智能体，可以更好指定搜索参数，如何处理搜索结果等
• 引用Skills、MCP等丰富大模型能力

了解到这些的时候，想着练习写一个Skills，实现网页内容抓取（其实很多东西都已经实现了，本文就是学习和分享），也了解一下Skills的开发

Skills的项目结构

skill-name/ ├── SKILL.md (唯一必需) │ ├── YAML 格式 (name, description 必须) │ └── Markdown instructions （介绍使用Markdown） └── Bundled Resources (可选的其他内容，和SKILL.md同级) ├── scripts/ - 存放可执行脚本（例如 Python 等） ├── references/ - 存放文档、API说明、领域知识 ├── examples/ - 存放示例文件 ├── evals/ - 存放测试说明 └── assets/ - 准备模板、图标、样板代码。确保格式正确（PPT、Word、图片等） 

SKILL.md元数据介绍

元数据字段:

scripts

Skills采用Prompt + Scripts架构，Scripts必须绑定特定运行时环境

• Skills的实现多采用Python脚本，也是大模型运行的主要环境
• Node.js Skills：需配置node_modules及package.json
• Bash Scripts：仅需基础Shell环境（但可能依赖系统工具包）

开发案例

Python还不太熟悉，用Node写了一个

主要实现的是：获取网页中所有的文本内容，如果有可以识别的媒体文件，将媒体资源的URL获取下来

整体项目目录结构，skills中website-content-fetch是本次的skill主目录，我们可以在一个项目中开发多个skill统一放在skills中，在一起维护也可以随时使用和优化其他skill

SKILL.md

--- name: website-content-fetch description: Fetch and extract content from websites. Use this skill whenever the user mentions fetching website content, extracting text from web pages, or needing to get content from a URL, even if they don't explicitly ask for a 'website content fetch' skill. --- 

package.json

{"name":"website-content-fetch","version":"1.0.0","description":"fetching website content","main":"scripts/fetch-content.js","scripts":{"test":"node scripts/fetch-content.js"},"keywords":["openclaw","skill","website","content","fetch"],"author":"","license":"MIT","dependencies":{"axios":"^1.6.2","cheerio":"^1.0.0-rc.12"}}

scripts

fetch-content.js

const axios =require("axios");const cheerio =require("cheerio");const path =require("path");const fs =require("fs");/** * Fetch website content * @param {string} url - The URL to fetch * @param {object} options - Optional parameters * @param {string} options.saveDir - Directory to save media files (optional) * @returns {Promise<object>} - The fetched content and metadata */asyncfunctionfetchWebsiteContent(url, options ={}){try{const response =await axios.get(url);const $ = cheerio.load(response.data);// Extract text contentlet content =$("body").text().trim();// Clean up content content = content.replace(/\s+/g," ");// Extract media resourcesconst media ={images:[],videos:[],audios:[],};// Extract images$("img").each((i, elem)=>{const src =$(elem).attr("src");const alt =$(elem).attr("alt")||"";if(src){const absoluteUrl =newURL(src, url).href; media.images.push({url: absoluteUrl,alt: alt,});}});// Extract videos$("video, iframe").each((i, elem)=>{let src =$(elem).attr("src");if(!src &&$(elem).attr("data-src")){ src =$(elem).attr("data-src");}if(src){const absoluteUrl =newURL(src, url).href; media.videos.push({url: absoluteUrl,});}});// Extract audios$("audio").each((i, elem)=>{const src =$(elem).attr("src");if(src){const absoluteUrl =newURL(src, url).href; media.audios.push({url: absoluteUrl,});}});// Save media files if saveDir is providedif(options.saveDir){// Create save directory if it doesn't existif(!fs.existsSync(options.saveDir)){ fs.mkdirSync(options.saveDir,{recursive:true});}// Save imagesfor(let i =0; i < media.images.length; i++){const image = media.images[i];try{const imageResponse =await axios.get(image.url,{responseType:"stream",});const imageName =`image_${i}_${path.basename(newURL(image.url).pathname)}`;const imagePath = path.join(options.saveDir, imageName);const writer = fs.createWriteStream(imagePath); imageResponse.data.pipe(writer);awaitnewPromise((resolve, reject)=>{ writer.on("finish",()=>resolve()); writer.on("error", reject);}); image.localPath = imagePath;}catch(error){ console.error(`Error saving image ${image.url}:`, error.message);}}// Save videosfor(let i =0; i < media.videos.length; i++){const video = media.videos[i];try{const videoResponse =await axios.get(video.url,{responseType:"stream",});const videoName =`video_${i}_${path.basename(newURL(video.url).pathname)}`;const videoPath = path.join(options.saveDir, videoName);const writer = fs.createWriteStream(videoPath); videoResponse.data.pipe(writer);awaitnewPromise((resolve, reject)=>{ writer.on("finish",()=>resolve()); writer.on("error", reject);}); video.localPath = videoPath;}catch(error){ console.error(`Error saving video ${video.url}:`, error.message);}}// Save audiosfor(let i =0; i < media.audios.length; i++){const audio = media.audios[i];try{const audioResponse =await axios.get(audio.url,{responseType:"stream",});const audioName =`audio_${i}_${path.basename(newURL(audio.url).pathname)}`;const audioPath = path.join(options.saveDir, audioName);const writer = fs.createWriteStream(audioPath); audioResponse.data.pipe(writer);awaitnewPromise((resolve, reject)=>{ writer.on("finish",()=>resolve()); writer.on("error", reject);}); audio.localPath = audioPath;}catch(error){ console.error(`Error saving audio ${audio.url}:`, error.message);}}}return{ content,length: content.length, url, media,};}catch(error){ console.error("Error fetching website content:", error);thrownewError(`Failed to fetch content from ${url}: ${error.message}`);}}// If run directly, test the functionif(require.main === module){const url = process.argv[2]||"https://example.com";const saveDir = process.argv[3];const options ={};if(saveDir){ options.saveDir = saveDir;}fetchWebsiteContent(url, options).then((result)=>{ console.log("Fetched content:"); console.log(`URL: ${result.url}`); console.log(`Length: ${result.length} characters`); console.log("Content:"); console.log(result.content);// Display media resources console.log("\nMedia resources:");if(result.media.images.length >0){ console.log("Images:"); result.media.images.forEach((image, index)=>{ console.log(`${index +1}. ${image.url} (alt: ${image.alt})`);if(image.localPath){ console.log(` Saved to: ${image.localPath}`);}});}if(result.media.videos.length >0){ console.log("\nVideos:"); result.media.videos.forEach((video, index)=>{ console.log(`${index +1}. ${video.url}`);if(video.localPath){ console.log(` Saved to: ${video.localPath}`);}});}if(result.media.audios.length >0){ console.log("\nAudios:"); result.media.audios.forEach((audio, index)=>{ console.log(`${index +1}. ${audio.url}`);if(audio.localPath){ console.log(` Saved to: ${audio.localPath}`);}});}}).catch((error)=>{ console.error("Error:", error.message);});} module.exports ={ fetchWebsiteContent };

evals

evals.json

{"skill_name":"website-content-fetch","evals":[{"id":1,"prompt":"Fetch content from https://example.com","expected_output":"Should return the text content of example.com","files":[]},{"id":2,"prompt":"Fetch content from a nonexistent domain","expected_output":"Should throw an error about failed to fetch content","files":[]},{"id":3,"prompt":"Fetch content without providing a URL","expected_output":"Should throw an error about URL being required","files":[]}]}

测试

将开发的skill目录website-content-fetch引入支持skill的AI助手或者开发IDE，此处不赘述，只要website-content-fetch放入对应的skills目录，助手或者IDE都可以识别到。

开发的时候，只需要在项目中，和AI工具说使用website-content-fetch skill为我抓取https://www.baidu.com中的内容即可，比如我使用Trae对话，AI会自动发现该工具，并识别scripts的语言，识别到node环境会开始install依赖，再用skill执行任务

install成功之后，环境准备就绪，就会按照skill开发的功能执行任务了

• 最后再反复测试结果，调试scripts功能，优化逻辑
• 丰富SKILL内容，为AI下次识别工具，对skill的功能和逻辑有更好的理解，更精确的执行任务

最后附上Github地址
aubrey-skills