WeChat Official Account 100k+ Popular Article Recommendations

Introduction

A delivery-type Skill for the scenario of "100k+ view hit articles from official WeChat accounts": pulls rankings by category and time window, provides a complete Markdown ranking in the conversation, and generates HTML that can be exported to PDF; supports subscription push by track (if enabled on the product side).

What can it do?

Overall Ranking / Track-specific Rankings: 23 standard categories + overall ranking; default TOP50 candidates, first-round conversation can preview 10 entries.
Data Integrity: Strongly relies on script stdout, Agent is prohibited from fabricating articles or view counts.
HTML + PDF: Ranking page follows WeChat's green visual specification, single-page PDF export (html2pdf.js).
Rule Review: Outputs "hit article rule analysis" based on real entries after the ranking, and asks whether to subscribe to the track.

Who is it for?

Official WeChat account editors / operators —— chase hot topics, check titles and account distribution
Content managers —— need shareable, printable ranking pages
Growth / business personnel —— monitor 100k+ content supply by track

Runtime Dependencies: Python 3; data scripts are declared in

scripts/fetch_hot_articles.py

in the repository (mainly standard libraries).

Functional Features

Core Functions

Ranking Pull:
```
scripts/fetch_hot_articles.py
```
→ write
```
temp_articles.json
```
→ output as-is in conversation.
Time Caliber: Synchronize previous day's data at 18:30 daily; Agent must calculate the query interval based on the current time (see reference documents).
HTML Generation:
```
scripts/generate_hot_html.py
```
, --display_count
must match the number of entries displayed in the current conversation.
Subscription Guidance: Ask whether to subscribe after rule analysis; do not execute subscription in the same round, wait for the user's next message to branch.

Highlight Features

Script output is the main content: Tables + details + statistics section cannot be deleted, modified or rearranged.
18:30 Watershed: Before and after 18:30, the "latest available date" is different, and fixed apology scripts must be used.
No Padding: Display fewer than 10 entries truthfully; if no data, follow the prompts in reference documents to switch to overall ranking or other categories.

Usage Scenarios

Load and strictly execute according to reference documents when users have the following demands:

User's Possible Questions	Agent's Action Summary
"Today's hit articles" "100k+ articles" "Latest hit articles"	Overall ranking intent + calculate date + `fetch_hot_articles.py` (usually preview 10)
"Technology / Finance / Health... 100k+ articles"	Map to standard `--type` + time interval + script
"What's popular recently / latest"	Default past 7 days interval (see time rule documents)
"Expand all / view 50 entries"	Change to `--mode full` in the same interval, synchronize HTML `display_count`

Typical Conversation Rhythm

Identify intent → 2. Run script and display as-is → 3. Hit article rule analysis → 4. Subscription inquiry → 5. Immediately generate HTML (do not wait for subscription reply) → 6. If user replies with subscription, proceed to subscription branch.

Important Data Notes

Update Rhythm: Ranking database aligns with the previous day at 18:30 daily; use 19:30 for external push scripts.
Time Inconsistency: Whenever the user's "today / yesterday" does not match the actual queryable interval, must use the original apology script in references/time-and-date-rules.md
.

Core Execution Rules (Must Follow)

Must run script first: Any ranking content comes from stdout of
```
fetch_hot_articles.py
```
, cannot be handwritten as a substitute.
Display stdout as-is: Paste the entire block including "data description, overview table, details, statistics".
Fixed six-step sequence: Cannot skip "rule analysis → subscription inquiry → HTML"; see references/agent-workflow.md
for details.
Consistent HTML entry count: If N entries are displayed in the conversation, use
```
generate_hot_html.py --display_count N
```
.
Detailed Drilling: Date intervals, parameter tables, output templates, API fields, etc., all use references as the single source of truth (see resource index in the next section).

Project Architecture

Directory (Reference)

text

wechat-10w-hot/
├── SKILL.md                              # This entry (simplified version)
├── scripts/
│   ├── fetch_hot_articles.py             # Pull + conversation Markdown output + temp JSON
│   └── generate_hot_html.py             # Read temp JSON → HTML
└── references/
    ├── agent-workflow.md                 # Six-step process, subscription, self-check
    ├── time-and-date-rules.md            # 18:30 and intervals, fixed scripts
    ├── script-parameters-and-output.md   # CLI, output format, four-dimensional analysis
    ├── html-pdf-visual-spec.md           # Visual, PDF, HTML parameters
    ├── usage-examples.md                 # Scenario-based command examples
    ├── api-spec.md                       # Interface field description
    └── category-mapping.md               # Category mapping

Data Flow (Conceptual)

text

User Request → Intent + Date Interval → fetch_hot_articles.py → stdout (conversation) + temp_articles.json
                                              ↓
                     generate_hot_html.py (display_count aligned) → HTML (PDF exportable)

Resource Index (Which to Read When)

File	When to Read
references/agent-workflow.md	Before executing any step: check sequence, subscription, self-check
references/time-and-date-rules.md	Calculate `start_date` / `end_date` , write apology scripts
references/script-parameters-and-output.md	Assemble CLI, understand stdout structure, four-dimensional content analysis
references/html-pdf-visual-spec.md	Generate HTML / PDF styles and commands
references/usage-examples.md	Compare overall ranking / field / full quantity / niche / no data
references/api-spec.md	Check URL, request parameters, `tenWReadingRank` structure
references/category-mapping.md	Map user's spoken language → standard `--type`

Frequently Asked Questions

Q1: Can I provide 10 sample articles without running the script? A: No. Without script output, it is considered that this Skill has not been executed.

Q2: User only wants HTML, not long text? A: Still must fully display the script stdout (per specification requirements), then generate HTML.

Q3:
temp_articles.json
has 50 entries but only 10 are discussed in the conversation, how many entries in HTML? A: 10 entries; use

--display_count 10

Q4: Where is the interface URL based on? A: Based on references/api-spec.md
and constants in

fetch_hot_articles.py

, if there is a discrepancy between the two, take the script as the standard and synchronize to correct the document.

wechat-10w-hot

NPX Install

Tags

SKILL.md Content (Chinese)