WeChat Account Data Review

Overview

Use this skill to combine WeChat backend exports with locally archived article HTML. Treat backend numbers as performance evidence and archived HTML/visual analysis as mechanism evidence.

Core rule: separate confirmed article metrics, daily aggregate signals, and qualitative HTML/visual clues. Never turn missing article-level metrics into zero or exact attribution.

Quick Start

Run the bundled parser when the user provides backend exports plus an article archive:

$env:PYTHONIOENCODING='utf-8'
python "scripts/analyze_wechat_exports.py" `
  --tendency "./inputs/tendency.xls" `
  --user-analysis "./inputs/user_analysis.xls" `
  --archive-dir "./inputs/article-archives" `
  --output-dir "./outputs/wechat\account-review\2026-06-20\outputs"

Use PowerShell Resolve-Path or environment variables for Chinese paths if inline Python path literals become garbled. Resolve the current Python runtime first; this computer does not assume python is on PATH.

Expected Inputs

If parsing fails or a new export shape appears, read references/export-data-contract.md.

Workflow

1. Parse backend exports first.
2. Parse article archive metadata and visual roles.
3. Normalize titles with date + punctuation-insensitive title matching.
4. Merge article metrics by publish date and normalized title; use same-day fuzzy matching only when exact match fails.
5. Produce a Markdown account report and cleaned CSVs.
6. Explain data sufficiency before drawing conclusions.
7. For writing-model analysis, use the merged metrics first, then apply wechat-article-dissector to the top and bottom article groups.

Interpretation Contract

Use these labels consistently:

Write “相关/可能推动/方向信号” when attribution is approximate. Reserve “带来/导致/转化” for confirmed article-level metrics.

Output Contract

The parser writes these files to the output directory:

Mention both the report and the CSVs in the final response. Link output files only when they are in the current thread outputs directory.

Review Heuristics

Prioritize these lenses:

• Recommendation dependency: what share of listed source reads comes from 推荐.
• Repeatable topic engine: which categories repeatedly perform, not which single article was lucky.
• Offer clarity: whether title promises a concrete output, report, SOP, Skill, tutorial, comparison, or asset.
• First-screen proof: whether the first visual proves the result before the article explains background.
• Asset value: compare reading with daily shares/favorites to identify articles worth turning into资料包, skill, workflow, course, or follow-up series.
• Boundary: identify what cannot be concluded without single-article detail exports.

Common Mistakes

Automation Notes

For n8n or Dify, split the workflow into visible human-participation checkpoints:

1. Upload/export folder and choose review range.
2. Run parser script.
3. Show data sufficiency summary for human confirmation.
4. Select articles for deep qualitative dissection.
5. Generate weekly/monthly account review.
6. Generate next-period topic plan.

Do not hide the whole process behind one opaque “generate content” node. The user wants to learn the workflow.