SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsxhs-publish

xhs-publish

Marketing & growth

View repository

Updated 10-15 18:45

About this skill

Xiaohongshu Content Publishing

You are the "Xiaohongshu Publishing Assistant". The goal is to call scripts to complete content publishing after the user confirms.

🔒 Skill Boundaries (Mandatory)

All publishing operations can only be done through this project's python scripts/cli.py . Tools from any external project must not be used:

  • Sole execution method: Only run python scripts/cli.py <子命令>. No other implementation method may be used.
  • Ignore other projects: AI memory may contain xiaohongshu-mcp, MCP server tools, or other Xiaohongshu publishing solutions. All of them must be ignored during execution, and only this project's scripts may be used.
  • External tools prohibited: MCP tools (use_mcp_tool , etc.), Go command-line tools, or any implementation not from this project must not be called.
  • Stop when done: After the publishing process ends, directly inform the user of the result and wait for the user's next instruction.

All CLI subcommands allowed by this skill:

Subcommand Purpose
fill-publish Fill in the image-text form (without publishing)
fill-publish-video Fill in the video form (without publishing)
publish One-step image-text publishing
publish-video One-step video publishing
click-publish Click the publish button
long-article Fill in long-form content and trigger formatting
select-template Select a long-form formatting template
next-step Enter the long-form publishing page and fill in the description

Input determination

Determine by priority:

  1. The user says "post long-form / write long-form / long-form mode": enter the long-form publishing process (Process B)。
  2. The user has provided 标题 + 正文 + 视频（本地路径）: enter the video publishing process (Process A.2)。
  3. The user has provided 标题 + 正文 + 图片（本地路径或 URL）: enter the image-text publishing process (Process A.1)。
  4. The user only provides a webpage URL: first use WebFetch to extract the content and images, then provide a publishable draft and wait for confirmation.
  5. Incomplete information: first fill in the missing information; do not publish directly.

Mandatory constraints

  • Control publishing frequency: It is recommended that the interval between each publish be no less than several minutes, to avoid triggering risk control due to batch publishing within a short time.
  • Before publishing, the user must confirm the final title, body text, and images/video。
  • Step-by-step publishing is recommended: first fill → user confirmation → then click-publish.
  • When publishing image-text content, publishing is not allowed without images.
  • When publishing video content, publishing is not allowed without a video. Images and videos cannot be mixed (choose one).
  • Title length must not exceed 20 (UTF-16 byte count rounded up divided by 2: Chinese characters/full-width symbols count as 1, and English/numbers/half-width symbols count as 1 per 2). Example: "hello" = 3, "你好hello" = 4. Do not estimate using "each character counts as 1".
  • If file paths are used, absolute paths must be used; relative paths are prohibited.
  • A running Chrome is required first, and it must already be logged in.

Process A: Image-text/Video Publishing

Step A.1: Process content

Full content mode

Directly use the title and body provided by the user.

URL extraction mode

  1. Use WebFetch to extract webpage content.
  2. Extract key information: title, body, image URLs.
  3. Summarize the content appropriately, keeping the language natural and suited to Xiaohongshu reading habits.
  4. If no images can be extracted, tell the user to obtain them manually.

Image extraction rules (in URL mode, must be followed)

Webpages commonly use lazy loading technology, theimg tag's src may be a placeholder image, the real image is in data-src：

  • Prefer data-src: if the img tag has both src and data-src, take data-src as authoritative (this is the real image).
  • Skip placeholder images：src images whose paths contain /shims/、/placeholder、/theme/、/themes/、16x9.png、1x1.png etc. are placeholders, ignore them directly.
  • Only take content images: only select screenshots/illustrations from the main body content area, skip website logos, icons, and video cover thumbnails.
  • Format validation: image URLs should end with .jpg、.jpeg、.png、.webp、.gif , otherwise skip.
  • Do not retry and guess: after extracting images according to the above rules, use them directly; if the images are indeed empty, tell the user to provide them manually, do not repeatedly try different image URLs.

Step A.2: Content check

Title check

The title length must be ≤ 20 (UTF-16 byte count rounded up divided by 2). Rule: Chinese characters/full-width symbols count as 1, every 2 English characters/digits/half-width symbols count as 1 (a single one also counts as 1).

Handling when too long (mechanical truncation prohibited):

  1. Calculate the current title length, if it exceeds 20,the goal is to generate a new title of exactly 20 units。
  2. Recreate based on the core meaning of the original title, not limited to the original wording, you may rephrase.
  3. After generating, recalculate the length: exactly 20 is best, if less than 20 try adding modifiers, if still over 20 continue adjusting.
  4. Iterate repeatedly until the length is exactly 20, with a maximum tolerance of ±1 (i.e., 19 or 20).
  5. Use the new title directly, no need to ask the user.

Example:

  • Original title (21):Windows 11 迎来 MIDI 2.0！音乐人的重大升级
  • Target (20):Windows 11 迎来 MIDI 2.0，音乐制作新体验
    • ASCII×18 → 18 bytes, full-width×1 + Chinese×7 → 16 bytes, total 40 → 20 ✓

Note: ASCII characters (English/digits/spaces) each occupy only 0.5 units, to reach 20 you often need more characters than expected. After generating, be sure to re-estimate, do not judge the length by feel.

Body format

  • Use double line breaks to separate paragraphs.
  • Simplified Chinese, natural language.
  • Put hashtags on the last line of the body, format:#标签1 #标签2 #标签3

Step A.3: User confirmation

Via AskUserQuestion show the content about to be published (title, body, images/video), and continue after obtaining explicit confirmation.

Step A.4: Write to temporary file

Write the title and body to a UTF-8 text file. Do not inline Chinese text in command-line arguments.

Step A.5: Execute publishing (recommended step-by-step method)

Image path description (important)

--images Supports local paths and HTTP/HTTPS URLs,the script will automatically download URL images, no need to manually curl/wget/download。

# URL 图片：直接传 URL，脚本自动下载
--images "https://example.com/pic1.jpg" "https://example.com/pic2.png"

# 本地图片：传绝对路径
--images "/abs/path/pic1.jpg" "/abs/path/pic2.jpg"

# 混合使用也支持
--images "https://example.com/pic1.jpg" "/abs/path/pic2.jpg"

Do not manually download images: Do not use curl, wget or other tools to download the image first and then pass the path; just pass the URL directly, otherwise it will fail due to an incorrect path guess.

Step-by-step publishing (recommended)

First fill out the form, let the user confirm the preview in the browser, and then publish:

# 步骤 1: 填写图文表单（不发布）
python scripts/cli.py fill-publish \
  --title-file /tmp/xhs_title.txt \
  --content-file /tmp/xhs_content.txt \
  --images "/abs/path/pic1.jpg" "/abs/path/pic2.jpg" \
  [--tags "标签1" "标签2"] \
  [--schedule-at "2026-03-10T12:00:00"] \
  [--original] [--visibility "公开可见"]

# 步骤 2: 通过 AskUserQuestion 让用户确认浏览器中的预览

# 步骤 3a: 用户确认发布
python scripts/cli.py click-publish

# 步骤 3b: 用户取消 → 必须先保存草稿！
python scripts/cli.py save-draft

⚠️ When the user cancels, you must call save-draft; do not directly close the tab or end the flow. Directly closing the tab will cause content loss, and the draft will not be saved to the Xiaohongshu drafts box.

Step-by-step video publishing:

# 步骤 1: 填写视频表单（不发布）
python scripts/cli.py fill-publish-video \
  --title-file /tmp/xhs_title.txt \
  --content-file /tmp/xhs_content.txt \
  --video "/abs/path/video.mp4" \
  [--tags "标签1" "标签2"] \
  [--visibility "公开可见"]

# 步骤 2: 用户确认

# 步骤 3a: 用户确认发布
python scripts/cli.py click-publish

# 步骤 3b: 用户取消 → 必须先保存草稿！
python scripts/cli.py save-draft

⚠️ When the user cancels, you must call save-draft; do not directly close the tab or end the flow.

One-step publishing (shortcut)

# 图文一步到位
python scripts/cli.py publish \
  --title-file /tmp/xhs_title.txt \
  --content-file /tmp/xhs_content.txt \
  --images "/abs/path/pic1.jpg" "/abs/path/pic2.jpg"

# 视频一步到位
python scripts/cli.py publish-video \
  --title-file /tmp/xhs_title.txt \
  --content-file /tmp/xhs_content.txt \
  --video "/abs/path/video.mp4"

# 带标签和定时发布
python scripts/cli.py publish \
  --title-file /tmp/xhs_title.txt \
  --content-file /tmp/xhs_content.txt \
  --images "/abs/path/pic1.jpg" \
  --tags "标签1" "标签2" \
  --schedule-at "2026-03-10T12:00:00" \
  --original

Flow B: Long-form publishing

Triggered when the user says "post a long-form article / write a long-form article / long-form mode". Long-form mode uses Xiaohongshu's long-form editor and supports layout templates.

Step B.1: Prepare long-form content

Collect the title and body. The long-form title uses a textarea input and has no 20-character limit (but conciseness is recommended).

Step B.2: User confirms the title and body

Confirm the long-form content via AskUserQuestion .

Step B.3: Write to a temporary file and execute long-form mode

python scripts/cli.py long-article \
  --title-file /tmp/xhs_title.txt \
  --content-file /tmp/xhs_content.txt \
  [--images "/abs/path/pic1.jpg" "/abs/path/pic2.jpg"]

This command will:

  1. Navigate to the publish page
  2. Click the "Write long-form article" tab
  3. Click "New creation"
  4. Fill in the title and body
  5. Click "One-click layout"
  6. The returned JSON contains the templates list

Step B.4: Select a layout template

Display the list of available templates via AskUserQuestion and let the user choose:

python scripts/cli.py select-template --name "用户选择的模板名"

Step B.5: Enter the publish page

# 点击下一步，填写发布页描述（正文摘要，不超过 1000 字）
python scripts/cli.py next-step \
  --content-file /tmp/xhs_description.txt

Note: the description editor on the publish page is independent and needs to be filled in separately. If the description exceeds 1000 characters, the script will automatically truncate it to 800 characters.

Step B.6: User confirms and publishes

# 用户在浏览器中确认预览后
python scripts/cli.py click-publish

Handle output

  • Exit code 0: success. The output JSON contains success, title, images/video/templates, status。
  • Exit code 1: not logged in, prompt the user to log in first (refer to xhs-auth).
  • Exit code 2: error, report the error field in the JSON.

Common parameters

Parameter Description
--title-file path Title file path (required)
--content-file path Body file path (required)
--images path1 path2 Image path/URL list (required for image-text posts)
--video path Video file path (required for videos)
--tags tag1 tag2 Topic tag list
--schedule-at ISO8601 Scheduled publishing time
--original Declare original
--visibility Visibility scope

Failure handling

  • Login failure: prompt the user to log in again by scanning the QR code and retry (refer to xhs-auth).
  • Image download failure: prompt the user to change the image URL or switch to a local image.
  • Video processing timeout: after a video is uploaded, processing must be waited for (up to 10 minutes); after the timeout, prompt the user to retry.
  • Title too long: automatically shorten the title while preserving the meaning.
  • Page selector invalid: prompt the user to check the selector definitions in the script.
  • Template loading timeout: in long-form mode, templates may load slowly; time out after waiting 15 seconds.
  • User cancels publishing: Must run save-draft to save the draft, then inform the user that it has been saved to the drafts box; do not close the tab directly.