x-ai-digest

Invoke / Invoke / Invocation

中文

"Use x-ai-digest to fetch today's AI hot topics"
"Use x-ai-digest to generate yesterday's X platform AI briefing"
"x-ai-digest help me check what AI-related activity there was on X the day before yesterday"
"Use x-ai-digest to organize recent AI tweets and provide reply suggestions"

English

"Use x-ai-digest to scrape today's AI hot topics from X"
"Use x-ai-digest to summarize AI posts from yesterday in English"
"Generate an X platform AI digest for 2026-01-20 using x-ai-digest"
"Use x-ai-digest to compile recent AI tweets with reply suggestions"

日本語

"Use x-ai-digest to summarize today's AI-related posts in Japanese"
"Use x-ai-digest to fetch yesterday's AI trends on the X platform"
"Use x-ai-digest to compile recent AI-related tweets"
"Use x-ai-digest to collect AI posts from the day before yesterday and create reply suggestions"

Objective

Scrape AI-related information for a specified date from the "For You" tab on the X platform homepage, and generate a structured daily briefing:

Deliverables: 1 Markdown briefing + 1 JSON raw data file + 1 WeChat official account share image
Each post: time, author, content, original link
Briefing content: hot topic summary, product launches, selected posts, reply suggestions
Share image: card-style design, including today's hot topic summary, suitable for sharing on WeChat official accounts/Moments
Default output:中文（English/Japanese content needs translation）

Prerequisites

dev-browser skill installed and running
- Need to use browser extension mode to connect to a browser already logged into X
- Startup command:cd skills/dev-browser && npm run start-extension
X account logged in
- The user needs to log in to their X account in the Chrome browser
- Ensure access to the "For You" tab

Input (ask first)

Parameter	Description	Default value
Date	Which day's data to scrape	Today (YYYY-MM-DD)
Quantity range	Target number of posts	50-100 posts
Output directory	Briefing save location	Current working directory
Whether to generate replies	Whether to generate reply suggestions for selected posts	Yes

Process

Step 1: Connect to the browser and visit the X homepage

// 使用 dev-browser 连接已登录的浏览器
const client = await connect();
const page = await client.page("x-home", { viewport: { width: 1920, height: 1080 } });

// 访问 X 主页
await page.goto("https://x.com/home");
await waitForPageLoad(page);

// 切换到「为你推荐」标签
await page.click('text=为你推荐');
await page.waitForTimeout(3000);

Step 2: Scroll and scrape posts

Scroll the page in a loop and extract for each post:

tweetId - Post unique ID
tweetUrl - Original link (https://x.com/{username}/status/{tweet_id})
datetime - Publication time in ISO 8601 format
username - Username
displayName - Display name
content - Post content

For detailed implementation, see references/scraping-guide.md

Step 3: Time filtering

Convert UTC time to Beijing time (UTC+8) for filtering:

// 北京时间 YYYY-MM-DD 对应的 UTC 时间范围
// 北京时间 00:00:00 = UTC 前一天 16:00:00
// 北京时间 23:59:59 = UTC 当天 15:59:59
const startUTC = new Date(`${date.slice(0, 4)}-${date.slice(5, 7)}-${parseInt(date.slice(8, 10)) - 1}T16:00:00.000Z`);
const endUTC = new Date(`${date}T15:59:59.999Z`);

Step 4: Deduplication and quantity control

Use tweetId for deduplication
Can stop when the cumulative total reaches 50 posts, with a maximum of 100 posts scraped
If 50 posts are still not reached after multiple scrolls, use the actual quantity
Stop when there is no new data after 10 consecutive scrolls

Step 5: Filter AI-related content

Use keyword matching to filter AI-related posts; see the keyword list in references/ai-keywords.md

Step 6: Generate the briefing

Use template assets/digest-template.md to generate the briefing, including:

Today's hot topics
Important product/feature releases
Industry observations
Selected posts and reply suggestions
Accounts worth following

Reply suggestion rules:

Generate replies in the original language of the post
Non-Chinese posts must include a Chinese explanation
Reply content must be valuable; avoid empty praise

Step 7: Generate the official account share image

Based on assets/share-card-template.html generate a share card:

Generate an HTML file: populate the template based on today's hot topics
Capture a screenshot to generate a PNG: use playwright to capture the .card element

import { chromium } from "playwright";

const browser = await chromium.launch({ headless: true });
const page = await browser.newPage({ viewport: { width: 900, height: 1800 } });

await page.goto(`file://${htmlPath}`);
await page.waitForTimeout(2000);

// 只截取 .card 元素，不要背景
const card = await page.$('.card');
await card.screenshot({
  path: `${outputDir}/ai-digest-${date}.png`,
  type: "png"
});

await browser.close();

The card content includes:

Title and date
Summaries of 3-4 hot topics today
Statistics (number of posts, number of topics, number of product releases)
Data source attribution

Guardrails

Only scrape publicly visible posts; do not attempt to bind login or bypass restrictions
Scroll interval >= 1 second to avoid triggering anti-scraping mechanisms
If a network error or page anomaly occurs, degrade to outputting the data already scraped
Confirm the path before outputting files to avoid overwriting existing files
Preserve original links when translating to ensure traceability

Output example

File structure

{output_dir}/
├── x-posts-{date}.json      # 原始数据
├── x-posts-{date}.md        # 帖子列表表格
├── AI简报-{date}.md         # 完整简报
└── ai-digest-{date}.png     # 公众号分享图片

Briefing structure

# AI 每日简报 | {date}

## 一、今日热点话题
### 🔥 话题1
### 🆕 话题2
...

## 二、精选帖子及回复建议
### 1. 作者 - 话题
**原文**：...
**链接**：...
**建议回复（语言）**：...
**中文说明**：...（如非中文）

## 三、今日金句
...

## 四、值得关注的账号
...

Resources

Scraping implementation guide:references/scraping-guide.md
AI keyword list:references/ai-keywords.md
Briefing template:assets/digest-template.md
Share card template:assets/share-card-template.html
Share card generation guide:references/share-card-guide.md