Xiaohongshu Operator

Xiaohongshu Operator

Automate Xiaohongshu (小红书) interactions through browser control. This skill uses the BrowserWing Executor HTTP API to drive a dedicated Brave browser. The user has a Brave installation reserved exclusively for this skill — it is not used for personal browsing.

Setup

The user has already logged in to Xiaohongshu in Brave. Login sessions persist in Brave's default profile, so no special data directory is needed.

To start the service:

bash scripts/start-browser.sh

Run from the xiaohongshu-operator/ skill directory.

Prerequisites

• BrowserWing Executor (npm install -g browserwing — auto-installed by the script)
• Brave Browser (dedicated for this skill, already logged in to Xiaohongshu)

How It Works

1. start-browser.sh launches BrowserWing with Brave on port 9222 (shared with x-com-operator).
2. BrowserWing reuses Brave's default profile, so the existing Xiaohongshu login session is available.
3. All API calls go to http://127.0.0.1:9222/api/v1/executor/.

Workflow

Before performing any Xiaohongshu operation, ensure the dedicated service is running:

bash scripts/start-browser.sh

Then follow the operation guides below. The API base for all operations is:

http://127.0.0.1:9222/api/v1/executor

General Pattern

Every operation follows the same rhythm:

1. Navigate to the right Xiaohongshu URL
2. Wait for page transitions or dynamic content to load (Xiaohongshu feeds can be slow)
3. Snapshot the page to understand its structure and get element RefIDs
4. Interact with elements using their RefIDs (@eX)
5. Verify the result by taking another snapshot or extracting content

CRITICAL NOTE ON SELECTORS: Xiaohongshu heavily obfuscates its CSS classes (e.g., .xg-feed-container, randomly generated classes). It rarely uses semantic id or data-testid attributes. Therefore, you MUST rely on the accessibility tree from GET /snapshot and match elements based on their inner text (e.g., "点赞", "收藏", "发布", "评论") or their structural position rather than hardcoded CSS selectors.

File Management

To keep the skill directory organized, follow these storage conventions:

• Clean Root: Do not save temporary files (snapshots, HTML dumps, screenshots) directly in the root xiaohongshu-operator/ folder.
• Use temp/ Directory: Always save intermediate debugging or data files in the temp/ subdirectory (e.g., temp/snapshot.json, temp/page.html).
• Automation Scripts: Store reusable automation scripts or recordings in the scripts/ folder.

Operations

1. Search (搜索)

Search for posts using a keyword.

# Navigate to search results using URL parameters directly
curl -X POST 'http://127.0.0.1:9222/api/v1/executor/navigate' \
  -H 'Content-Type: application/json' \
  -d '{"url": "https://www.xiaohongshu.com/search_result?keyword=SEARCH_TERM_HERE"}'

# Wait for content to load (the feed usually takes a moment)
sleep 3
# Or use the wait API to look for a known text string if applicable, e.g.:
# curl -X POST 'http://127.0.0.1:9222/api/v1/executor/wait' ...

# Snapshot to see the loaded results
curl -X GET 'http://127.0.0.1:9222/api/v1/executor/snapshot'

# Note: Sometimes titles/text on feed pages aren't cleanly mapped in the accessibility tree.
# You can use the extract API to get all post titles directly:
curl -X POST 'http://127.0.0.1:9222/api/v1/executor/extract' \
  -H 'Content-Type: application/json' \
  -d '{"selector": "a.title", "multiple": true, "fields": ["text", "href"]}'

To load more results, you can scroll down:

curl -X POST 'http://127.0.0.1:9222/api/v1/executor/scroll-to-bottom' \
  -H 'Content-Type: application/json'

2. View a Post & Comments (查看帖子与评论)

When you want to view a post detail and its comments, navigate to its URL (usually starts with https://www.xiaohongshu.com/explore/...).

# Navigate to the post
curl -X POST 'http://127.0.0.1:9222/api/v1/executor/navigate' \
  -H 'Content-Type: application/json' \
  -d '{"url": "https://www.xiaohongshu.com/explore/POST_ID"}'

# Wait for the post and comments to load
sleep 3

# Snapshot the page
curl -X GET 'http://127.0.0.1:9222/api/v1/executor/snapshot'

The snapshot will show the post content, images/video elements, and the comments section below or to the side. You can extract the text from the snapshot.

3. Like (点赞) and Collect (收藏)

To like or collect a post, first navigate to its detail page (or identify it in the feed) and take a snapshot.

# Snapshot to find the Like or Collect button RefID
curl -X GET 'http://127.0.0.1:9222/api/v1/executor/snapshot'

# Look for elements in the snapshot whose accessible name indicates "Like" (often the count of likes) 
# or a button that represents liking/collecting. Wait for it if not immediately present.

# Click the element (replace @eXXX with actual RefID)
curl -X POST 'http://127.0.0.1:9222/api/v1/executor/click' \
  -H 'Content-Type: application/json' \
  -d '{"identifier": "@eXXX"}'

Because CSS classes are obfuscated, finding the exact like/collect button may require looking at sibling elements of the post content or checking elements near the comment input box in the snapshot.

4. Post (发帖)

[!WARNING] Xiaohongshu's Creator Center (https://creator.xiaohongshu.com/publish/publish) uses a complex Micro-Frontend (Qiankun) rendering architecture. Standard file input elements are deeply hidden. The most reliable automated posting method is to use Xiaohongshu's built-in "Text-to-Image" (文字配图) feature.

Step 1: Navigate to Creator Center

curl -X POST 'http://127.0.0.1:9222/api/v1/executor/navigate' \
  -H 'Content-Type: application/json' \
  -d '{"url": "https://creator.xiaohongshu.com/publish/publish?source=official"}'

There are two main ways to post: Text-to-Image (Automated) and Standard Image Upload (Manual Intervention Required).

Option A: Text-to-Image (Fully Automated)

1. Click "上传图文" (Upload Graphic/Text) -> "文字配图" (Text to Image)

# Click "上传图文"
curl -X POST 'http://127.0.0.1:9222/api/v1/executor/evaluate' \
  -H 'Content-Type: application/json' \
  -d '{"script": "() => { Array.from(document.querySelectorAll(\"span\")).find(el => el.textContent === \"上传图文\")?.click(); return true; }"}'

sleep 1

# Click "文字配图"
curl -X POST 'http://127.0.0.1:9222/api/v1/executor/evaluate' \
  -H 'Content-Type: application/json' \
  -d '{"script": "() => { Array.from(document.querySelectorAll(\"span\")).find(el => el.textContent === \"文字配图\")?.click(); return true; }"}'

2. Generate Image from Text

# Type the image text into the generator input
curl -X POST 'http://127.0.0.1:9222/api/v1/executor/type' \
  -H 'Content-Type: application/json' \
  -d '{"identifier": "div[role=\"textbox\"]", "text": "自动生成的配图文字"}'

# Click "生成图片" (Generate Image)
curl -X POST 'http://127.0.0.1:9222/api/v1/executor/evaluate' \
  -H 'Content-Type: application/json' \
  -d '{"script": "() => { document.querySelector(\"div.edit-text-button\").click(); return true; }"}'
  
# Wait 2-3 seconds for generation, then click "下一步" (Next)
sleep 3
curl -X POST 'http://127.0.0.1:9222/api/v1/executor/evaluate' \
  -H 'Content-Type: application/json' \
  -d '{"script": "() => { Array.from(document.querySelectorAll(\"button\")).find(el => el.textContent.includes(\"下一步\"))?.click(); return true; }"}'

3. Enter Posting Title and Content

# Type Title
curl -X POST 'http://127.0.0.1:9222/api/v1/executor/type' \
  -H 'Content-Type: application/json' \
  -d '{"identifier": "input[placeholder=\"填写标题会有更多赞哦\"]", "text": "这里是标题"}'

# Type Content (with tags)
curl -X POST 'http://127.0.0.1:9222/api/v1/executor/type' \
  -H 'Content-Type: application/json' \
  -d '{"identifier": "div[role=\"textbox\"]", "text": "你的帖子正文内容 #话题[话题]# "}'

4. Click Publish (发布)

curl -X POST 'http://127.0.0.1:9222/api/v1/executor/evaluate' \
  -H 'Content-Type: application/json' \
  -d '{"script": "() => { Array.from(document.querySelectorAll(\"span\")).find(el => el.textContent === \"发布\")?.click(); return true; }"}'

Option B: Standard Image Upload (Manual Intervention)

If you need to upload specific local images:

1. Upload: Rely on the user to manually click "上传图文" on the page and select the media file from their computer.
2. Title/Content: Once the user is on the text editor screen, use the snapshot API to locate the textboxes, or use the type commands from Option A.
3. Publish: Use the evaluate script above to find and click "发布" or "发布笔记".

5. View Personal Pages (查看个人主页)

# Navigate to a user's profile
curl -X POST 'http://127.0.0.1:9222/api/v1/executor/navigate' \
  -H 'Content-Type: application/json' \
  -d '{"url": "https://www.xiaohongshu.com/user/profile/USER_ID"}'

sleep 3

# Snapshot to view their posts
curl -X GET 'http://127.0.0.1:9222/api/v1/executor/snapshot'

Tips for Success

• Wait and Retry: Xiaohongshu uses lazy loading heavily. Elements might not be in the DOM until you scroll or wait a few seconds.
• Login States: If a page continuously redirects to a login portal (e.g., /user/login), notify the user that they might need to log in to Xiaohongshu manually in the dedicated Brave browser session.
• DOM Inspection: Use evaluate to extract parts of the DOM if the accessibility tree doesn't provide enough context to distinguish between like buttons vs other icons.

curl -X POST 'http://127.0.0.1:9222/api/v1/executor/evaluate' \
  -H 'Content-Type: application/json' \
  -d '{"script": "document.body.innerHTML.substring(0, 1000)"}'