Before executing the complete process, you must read and follow references/core_workflow.md the rules, date calculations, and script invocation instructions in this document; this article retains the outline and hard constraints, and the detailed rules are subject to that document.
Introduction
For creators and operators who need to improve title click-through rates and seeding efficiency on theXiaohongshuside: align the ideas, products, or topics given by users with the characteristics of similar viral content, then produce actionable title plans.
Through this Skill, you can:
- 📊 Data alignment: obtain popular note samples by keyword (see the script and workflow for details), avoiding fabricated interactions and cases out of thin air
- 🔍 Analyze first, then write: you must first complete the "Viral Title Analysis" before generating new titles, ensuring the patterns are traceable
- ✍️ Batch delivery: output 10 titles at a time in a unified format, facilitating A/B testing and selection
Suitable for new bloggers starting accounts, brand/e-commerce operators writing note titles, MCN batch proposals, etc.Environment: must be able to run scripts/fetch_xhs_trends.py(Python, dependencies see frontmatter).
Features
Core Features
- Keyword-driven: supports user-specified product terms, topics, or generalized categories; generalized terms go through the expansion confirmation process (see Usage Guide)
- Time strategy: when data is insufficient, only automatically expand the time window (recent 1→3→7→30 days),do not change keywords
- Two-stage output: analyze viral title structures, audiences, and patterns → then create 10 new titles
- Fixed delivery format: match index, reference viral links, detailed recommendation reasons, format see "Output Format" below
Highlights
- Clear form boundaries: only search the product form entered by the user, do not arbitrarily expand to adjacent categories (e.g., cream ≠ spray)
- Compliant wording: the terms "crawl" and "scrape" are prohibited; uniformly say "obtain"; output should avoid vulgar and non-compliant expressions, and sensitive original titles need neutral summarization
Usage Guide
Basic Path (Model Side)
Step 1: Understand the input and extract keywords
Users may say a single word, a short phrase, or a requirement description;uniformly understand it as: generate Xiaohongshu viral titles based on that content. If the information is extremely vague, you may ask "What is the core product or topic" and "Is a time range specified";do notswitch tracks for queries without authorization before confirming the generalized term strategy.
Step 2: Generalized term branch (only when the input is a broad category term)
You must first output subdivided direction recommendations andstop and waitfor the user to reply "expand" or "do not expand", then call the script. It is forbidden to execute the query in the same round of dialogue without waiting for a reply.
User: Help me come up with a few titles for the beauty direction. Assistant: Identified as a generic term. Recommended sub-directions: ... Please reply "Expand" to also search the following terms, or "Do not expand" to search only "beauty".
Step 3: Time range and data acquisition
According to references/core_workflow.md , convert the user's colloquial date into startDate（yyyy-MM-dd), execute:
python scripts/fetch_xhs_trends.py --keyword "<关键词>" --start-date <yyyy-MM-dd>
Separate multiple keywords with English commas; for site-wide trending use --keyword "". When data is insufficient, only extend the time, do not change the keywords, and explain to the user that it has been extended to the past Y days.
Step 4: Read the data report
Read the generated 关键词_爆款数据.md(or the path specified by the workflow), to analyze and cite real photoId links.
Step 5: Analyze first, then generate titles
You must first output the "Viral Title Analysis" structural block, then according to theoutput formatprovide 10 new titles; the new titles are innovations based on patterns, not copies of the original titles.
Step 6: Self-check the format
Check item by item the title line, match index (8–10, one decimal place), reference link form, recommendation reason, and separator line, consistent with the workflow verification checklist.
Command quick reference
| Command example | Function |
|---|---|
python scripts/fetch_xhs_trends.py --keyword "防晒霜" --start-date 2026-04-01 |
Acquire data by keyword and start date and write a Markdown report |
python scripts/fetch_xhs_trends.py --keyword "a,b,c" --start-date 2026-04-01 |
Multiple keywords (comma-separated) |
python scripts/fetch_xhs_trends.py --keyword "" --start-date 2026-04-01 |
No track, check site-wide trending |
On failure or insufficient data: explain the reason (API failure, insufficient number of items within the time window, etc.),must not fabricatenote titles, interaction counts, and links; if it is still insufficient after extending to the past 30 days, present it truthfully and indicate that data is limited.
Key rules (consistent with core_workflow, must not be violated)
- Execution order: You must first "analyze viral titles", then "generate new viral titles"; the order must not be reversed.
- Insufficient data: prioritize automatically extending the time (past 1 day → past 3 days → past 7 days → past 30 days),changing keywords is absolutely prohibited。
- Generic term handling: first output sub-term recommendations and wait for the user's choice (reply "Expand" or "Do not expand"), then call the script to query.
- Precise keywords: query only the product the user entered, and do not expand to related but different product forms.
- Wording standards: the use of "crawl" and "scrape" is prohibited; uniformly use "acquire"; the analysis and titles visible to the user must be compliant, avoiding vulgar, insulting, illegal, or non-compliant expressions; if the reference data contains sensitive original titles, summarize them neutrally and do not repeat sensitive words for the sake of attracting attention.
Output format
Each recommended title must be output strictly in the following format:
**标题X：[标题内容]**（20字以内）
📈匹配指数：[8-10之间，保留一位小数]
🔥参考的爆款：[标题文字](https://www.xiaohongshu.com/explore/{photoId})（互动数：XXX）
👍推荐理由：[详细推荐理由]
---
Output 10 titles in total, with each title block separated by --- .
Use cases
| Scenario | Role | Requirement description | How to use |
|---|---|---|---|
| Titling for new products | E-commerce/brand operations | The product has selling points but you do not know how to write titles in the Xiaohongshu style | Provide product terms and selling points; pull data; analyze viral titles first, then produce 10 alternatives |
| Daily account updates | Blogger / director | Same track lacks topic selection and title formulas | Enter a track word or niche word; iterate the title library according to data patterns |
| Category trial | Novice creators | Only think of broad categories such as "beauty" and "fashion" | Use generalized word recommendations; after the user chooses whether to expand or not, query again |
| Site-wide trend chasing | Operations | Want to see recent site-wide popular content and then adapt it | Leave keywords empty and query the whole site according to the workflow; then narrow down to specific creation |
Notes and boundaries
- Data: comes from the script-agreed interface and stored snapshots, not real-time; interaction counts are subject to the report,fabrication prohibitedReference notes and
photoId。 - Task ownership: the entire process of this Skill is executed by themain agent . Do not split it into sub-agents, to avoid breaking the format and order.
- Compliance: comply with community guidelines and advertising compliance practices; reference links must be traceable, and sensitive content in the analysis should only be described neutrally.
- Scope: does not promise to replace platform review results; content published on the user side still requires its own compliance review.
Resource index
- Complete process and verification checklist:references/core_workflow.md
- Data query script:scripts/fetch_xhs_trends.py