Introduction
A data insight Skill for WeChat Official Account topic selection and creation: continuously collects viral articles across the web, filtering by dimensions such as "low follower count but high reads, top reads, data trending upward, and top original content," providing account owners with referenceable viral samples.
What can it do?
- 🔍 Keyword discovery: Query the most matching WeChat Official Account viral content by track/niche term
- 📊 Smart filtering: Combining user intent and content relevance, select up to 10 high-quality recommendations from candidates
- 🃏 Dual-format display: Markdown cards + HTML grid cards, easy to read and share
- 🧭 Niche guidance: After displaying results, proactively provide 10 niche track terms for further exploration
Who is it for?
- WeChat Official Account operators / editors — find topics, see benchmarks
- Content creators — understand recent high-heat topics and title patterns
- MCN / growth — observe viral characteristics of a vertical in batches
Runtime dependencies：Python requests>=2.28.0。
Features
Core features
- Viral article discovery: Search trending WeChat Official Account articles across the entire site
- Time window: Defaults to the last 7 days; can query WeChat Official Account viral data from up to the last 30 days
- Broad term handling: After identifying a broad category term, must first provide niche recommendations and wait for the user's "expand / do not expand"; running scripts in the same turn is prohibited
- Output artifacts: Script generates
{keyword}_爆款数据.html; filtering logic can refer to{keyword}_爆款数据.json(if produced by the script)
Highlights
- Intent first: Prioritize extracting niche direction terms from the user's description, avoiding simply searching with overly broad category terms like "workplace" or "emotion"
- Data honesty: Not a real-time snapshot; requests such as today / over 30 days have fixed wording to explain the boundaries
- No padding: Up to 10 items; if fewer than 10, display truthfully; fabrication is prohibited
- Display loop: Each item includes a recommendation reason (≥15 characters, no empty template phrases); at the end must attach 10 niche terms to guide the next query
Use cases
When users raise the following types of needs, this Skill should be invoked:
| Users may ask | Agent behavior summary |
|---|---|
| "What viral articles are there recently" "Show me what's trending across the site" | Pass keyword "", query site-wide trending by default time window |
| "What viral articles are there in the workplace / emotion field" (broad category term) | Follow the broad term process: first provide niche terms + wait for "expand / do not expand" |
| "Workplace communication skills" "Parent-child education topics" | Treat as niche terms, can directly invoke the script (no expansion inquiry needed) |
| "xxx from the last 15 days" | Map colloquial phrasing to startDate = 今天 - N 天 then query |
| "Want high-like / high-share ones" | The intent type is classified as data-requirement type; when filtering, focus on interaction and data performance |
Typical examples
Scenario 1: No track browsing
User: What popular articles has the official account had recently? Assistant:
--keyword ""Pull all-site popular items → display cards and HTML → then give 10 popular track subcategory terms.
Scenario 2: Broad category terms need expansion
User: Help me look at popular workplace-related content. Assistant: Identify "workplace" as a generalized term → reply only with a list of subcategory terms and explain "expand / do not expand" → do not call the script in this round → execute the query only after the user replies.
Scenario 3: Comes with detailed subcategory description
User: I run an emotional content account, covering dating, marriage, and mother-in-law/daughter-in-law topics; give me popular references. Assistant: Extract multiple terms such as "emotions, dating tips, marriage management, mother-in-law/daughter-in-law relationship" into a comma-separated query → intent filtering → display ≤10 items.
Important data notes
Time and real-time nature
- Data time lag explanation: Article interaction data is as of data retrieval time, interaction volume may continue to grow and will continue to be updated later.
- Default time range:last 7 days（
startDate = 今天 - 7 天); when data is insufficient, automatically expand the range, and use the fixed wording below to inform the user。
Boundary wording (must be used verbatim)
- When the user mentions "today":"We are very sorry, today's data has not been updated yet; we have shown you the most recently available data"
- When the time requested by the user exceeds 30 days:"We are very sorry, currently only data from the last 30 days is supported; we have shown you the closest data"
Core execution rules (must be followed)
- Generalized terms must be asked about before querying: when identified as a generalized term, calling the script directly is prohibited; first output subcategory term recommendations and wait for the user to choose.
- Correct order: keyword extraction → determine whether it is a generalized term → if it is a generalized term, ask → call the script only after the user replies。
- Mandatory wait: after outputting subcategory recommendations, script execution must stop in this round; the script must not continue to be called within the same assistant turn.
Generalized terms vs. specific terms (definitions must be retained for determination)
- General term: A summary term with a high level of abstraction and broad coverage, without specific scenario/attribute modifiers, often a major industry category; its semantics are a superordinate concept (e.g., "workplace" includes work/workers; "emotions" includes dating/marriage; such as AI); in context it is often paired with "field", "type", etc. (e.g., "workplace field", "emotion type").
- Specific term: A term with a low level of abstraction and clear reference, containing scenario/attribute modifiers, and a direct subclass of a generalized term; its semantics are a subordinate concept (e.g., "workplace communication" belongs to the workplace skills subclass; "dating tips" belongs to the emotions subclass); its structure often contains modifiers (e.g., "mother-in-law/daughter-in-law" → "mother-in-law/daughter-in-law relationship"; "parent-child" → "parent-child education").
Project architecture
Directory and scripts (for reference)
gzh-explosive-content-detector/
├── SKILL.md # 本技能文档
├── scripts/
│ └── fetch_gzh_trends.py # 数据拉取与 HTML 生成入口
└── （运行产出）{keyword}_爆款数据.html
（若脚本生成）{keyword}_爆款数据.json
Data flow (concept)
用户请求 → 意图/关键词/泛化词判定 → fetch_gzh_trends.py → HTML（+ JSON）
↓
筛选与卡片渲染说明 ← 意图 + 相关性 + 数据分
Common Q&A
Q1: If the user only says "workplace", can we query directly? A：Cannot. You must first go through generic term expansion and wait for "Expand / Do not expand"; scripts are prohibited in the same round.
Q2: When should the keyword be left empty?
A: When the user has not provided any track or topic term, and only generally wants to see viral / trending content, pass "" to search the entire site.
Q3: What should be done if the data is empty? A: Indicate that the keyword is relatively niche, and suggest replacing it with a more popular track keyword;must notswitch to unrelated keywords or mistakenly trigger the generic term expansion process just because there is little data.
Update log
v1.1.1
- Alignment verification with the original execution logic: restore fixed phrasing, filtering steps, card examples, the original self-check text, and constraints such as "expansion = comma-separated multiple terms", to avoid ambiguity introduced by refactoring
v1.1.0
- Restructure the document according to the template (Introduction / Scenarios / Guide / Architecture / FAQ)
- Merge duplicate data descriptions and display requirements, and correct the formatting of the generic term example code block
v1.0.0
- Viral content detection, generic term governance, filtering and dual-form display, niche recommendations, and self-check checklist