Write Xiaohongshu (XHS: Research First, Write Then Publish)

Quick Start (Minimal Usage)

When the user provides a topic, directly run the full process from "Step 0→6"; finally output only:

Title (≤20 characters)
Body text (≤1000 characters)
Hashtags (#xxx #yyy…)
Images (1~2 images, 9:16 aspect ratio preferred)
Publishing result (if Xiaohongshu MCP is available)

Hard Limits (Must Be Followed)

Title: ≤ 20 characters
Body text: ≤ 1000 characters

Counting standard: Count by "characters" (including spaces, punctuation, line breaks, hashtags, etc.). To be safe: keep titles ≤18 characters and body text ≤950 characters to leave room for hashtags and formatting.

Required Output Components

The output of this skill is divided into two sections:

Analysis Summary Report (Must be provided after completing Step 1+2 analysis, structure follows the template below)
Final Publishable Draft (Title + Body text + Hashtags + Images + Publishing result)

text

Title: <≤20 characters>
Body Text:
<≤1000 characters>
Hashtags: #tag1 #tag2 #tag3 ...
Images: <1~2 images, 9:16 preferred; provide image URL or local path>

If the publishing interface supports a separate "hashtag field", do not include hashtags in the body text; place hashtags in the publishing parameters. Ensure the body text itself still meets the ≤1000 character limit.

Full Process (Must Be Followed in Order)

mermaid

flowchart TD
  A[Check Login Status] --> B[Find Top 10 Image Posts on Xiaohongshu]
  B --> C[Analyze Title & Body Patterns + Reasons for High Likes]
  C --> D[Capture Comments & Analyze Emotional Resonance Points]
  D --> E[Supplement Background & Fact-Check via Firecrawl]
  E --> F[Write Title/Body/Hashtags + Word Count Self-Check]
  F --> G{Did User Provide Images?}
  G -- Yes --> H[Use User-Provided Images]
  G -- No --> I[Find 1~2 High-Definition 9:16 Images Online]
  H --> J[Publish via Xiaohongshu MCP]
  I --> J

Step 0: First Check Login Status

First call Xiaohongshu MCP to check login status. If not logged in, provide a QR code for the user to log in.

Step 1: Find Top 10 "High-Like" Relevant Image Posts on Xiaohongshu and Summarize Patterns

Goal: First identify the most popular writing styles for similar content before starting to write.

Mandatory Threshold (Must Execute)

Successfully obtain at least 10 "complete and usable" image posts before proceeding to Steps 2/3/4/5/6.

Requirements:

Only select image posts (exclude videos/live stream clips)
Prioritize relevance, then high like counts; filter for "most likes"
Filter for posts from the past 6 months
Only use Xiaohongshu MCP's get_specified_post tool to retrieve posts, filter for "image" type, "past 6 months", retrieve 20 posts (actual results may be less than 20, tolerate down to 10 posts). Do not call other MCP tools; this tool returns all required information.

Record these fields for each top post (for subsequent analysis):

Title (original text)
Likes/Collections/Comments (if available)
Body structure (summarize in 1 line: e.g., "Conclusion → Checklist → Steps → Feedback → Interaction")
Key hook (the "attraction point" in the first sentence)
Evidence type (experience, comparison, data, screenshots, before-after changes…)
Interaction design (questions, polls, comment keywords, giveaways, collection guides…)

Output a "Pattern Summary" (directly reusable for writing):

Title patterns (length, sentence structure, keywords, symbols/numbers, emotional words, benefit points)
Body text patterns (paragraph rhythm, checklists/steps, transitions, pitfall avoidance, ending CTA, tone style)
Reasons for high likes (at least 3 points, as specific as possible to "writing style/information density/resonance points/executability/tone style")

Step 2: Analyze Reasons for "Emotional Resonance" Based on Comments from Step 1 Data

Goal: Identify why audiences "want to comment/forward/collect", summarize 3-6 resonance point conclusions and 1 interaction design suggestion.

Methodology:

Categorize "resonance triggers" in comments, such as:
- Shared pain points (Me too…)
- Identity recognition (Office workers/students/moms…)
- Anxiety and solutions (Finally a way/saved)
- Values (Self-discipline, relaxed vibe, cost-effectiveness, minimalism…)
- Contrast/surprise (Turns out this works too)
- Replicability (I can do this too)
- Desire to share (I want to try this too)
- Curiosity (What is this?)
- Resonance points (I do this too)

Output:

3～6 "resonance point conclusions" (each with 1～2 original comment snippets/summaries)
1 "interaction design suggestion" (e.g., what to ask at the end to encourage comments)

Analysis Summary Report (Must Output, Then Automatically Proceed to Subsequent Steps)

After completing Steps 1+2 (and meeting the "usable samples ≥10" requirement), must first output a summary report following the structure below (content style/granularity refer to user-provided examples), then automatically proceed to Steps 3→6 in the same output.

If the sample threshold is not met: Stop directly per Step 1's mandatory threshold, do not proceed to Steps 2/3/4/5/6, and do not fabricate a report.

Report Template:

text

[Analysis Summary Report]
Comment Count/Interaction Rank (From Highest to Lowest)
1. <Title> - <Comment Count> (<Image/Video>)
2. ...
...
10. ...
11. ...

Title Pattern Analysis
- <Pattern 1>
- <Pattern 2>
- <Pattern 3>
...

Content Pattern Analysis
- <Pattern 1>
- <Pattern 2>
...

Cover Pattern Analysis
- <Pattern 1>
- <Pattern 2>
...

Summary
Three Core Elements of Viral Posts:
1. <Element 1>
2. <Element 2>
3. <Element 3>

Interaction Catalysts:
- <Catalyst 1>
- <Catalyst 2>
...

Step 3: Supplement Background Knowledge via Firecrawl MCP (And Fact-Check)

Goal: Avoid empty content, errors/exaggerations; supplement "credible information".

Methodology:

Expand search with topic keywords: concept explanations, common misconceptions, authoritative suggestions, data (preferably from credible sources)
Refine into "5～10 quotable background points" (short sentences suffice)
Downplay claims that are easily questioned (efficacy, benefits, comparison conclusions):
- Rewrite to "My personal experience/suitable for some people/may be effective"
- Or add conditions (target audience, prerequisites)

Output a "Background Knowledge Card":

Key conclusions (5～10 points)
Risks/Notes (2～4 points)

Step 4: Write Title + Body Text + Hashtags Based on Above Analysis (And Mandatory Word Count Self-Check)

Writing Requirements (Xiaohongshu-style, but authentic):

Colloquial, short sentences, line breaks every 1–3 sentences
High information density: fewer empty words, more "actionable steps/checklists"
Appropriately use emojis (to enhance rhythm/emotion), but do not overuse
Natural ending, written like a real user, avoid "marketing account tone"

Anti-AI Tone Guidelines (Must Execute)

Write more "I/When I/I found/The pitfalls I stepped on", use less "everyone/users/audiences/the following suggestions"
Include 2～3 specific details: time points, scenarios, before-after comparisons, a specific line of dialogue/a shot/a small action
Allow imperfection: Use "I feel/I guess/maybe/Not sure but…" (don't pretend to be authoritative)
Avoid template phrases: Avoid "In general/To sum up/Therefore/First second last/Don't miss it/Hurry up"
Talk like chatting with a friend: Short sentences, line breaks, occasional colloquial words (but don't overdo it)

Ending Interaction (More Authentic, Avoid Marketing Account Tone)

Forbidden Ending Phrases (Do Not Use):

"Leave your opinion in the comments/What do you think sisters/Follow me/Like collect forward/Private message me to get…"
"You must watch/Regret not watching/The most comprehensive on the whole network/Suggest collecting and re-watching"

Recommended Ending Styles (Choose 1, as specific as possible):

"I'm stuck on <specific problem> right now, how do you think I should handle it?"
"Do you stand <A/B> or <C/D>? I'm a bit confused."
"Is anyone else who was directly heartbroken by <a certain shot/a certain line>…"
"I want to recheck <a certain point>, did anyone else notice it too?"

Default Structure (Stable and Universal):

1 sentence conclusion/result (who it benefits)
3 key points (checklist/steps/pitfall avoidance)
1 paragraph explanation (why do this)
1 interactive CTA

Word Count Gatekeeper (Must Do Last):

If title character count >20: Delete redundant words, remove subtitles, rewrite to short sentences, remove unnecessary symbols
If body text character count >1000: First delete repetitions and filler words, then merge sentences, then compress steps/reduce cases
Only output the compliant version in the end

Step 5: If User Did Not Provide Images, Find 1～2 High-Definition Images Online (9:16 Preferred)

Requirements:

Only 1～2 images needed, high-definition, clean, strongly relevant to the topic
Priority aspect ratio: 9:16 (vertical image)
Prioritize commercial-use/royalty-free sources (e.g., Pexels / Unsplash, etc.)
When outputting, provide: Image URL (and optionally: local download path)

Step 6: Publish via Xiaohongshu MCP

When publishing (adapt to MCP capabilities, do as much as possible):

Upload 1～2 images (prefer vertical images)
Set title, body text, hashtags
If supported: First create a draft → Echo draft information → Then publish

After successful publishing, output:

Published link/ID (if available)
Publishing parameter summary (title/hashtags/number of images)

Fallback Strategy (When Tools Are Unavailable)

If Firecrawl MCP or Xiaohongshu MCP is not available in the current environment:

Still complete the "analysis → writing → image link" process, but skip automatic publishing
List the shortest checklist of information the user needs to supplement (e.g., provide 3~5 benchmark post links, or authorization/interface configuration)

Examples

Example 1: User Only Provides Topic

User Input:

Write a Xiaohongshu image post: Topic is "Quick Dinner for Office Workers", suitable for single people, budget ≤20 yuan.

What You Need to Do:

Use topic keywords to find "image posts" on Xiaohongshu, summarize title/body text patterns
Capture comments from these posts, extract resonance points and interaction questions
Use Firecrawl to supplement background information related to "health/meal prep/safety"
Output title + body text + hashtags, and perform word count self-check
If no images are provided, find 1~2 high-definition 9:16 vertical images
Publish via Xiaohongshu MCP (if available)

Example 2: User Provides Images