WeChat Favorites Knowledge Base

Problem

WeChat favorites often accumulate into a backlog where tags are inconsistent, duplicates are common, and reading status is unclear. This skill assumes a parsed favorite.db export and treats favorites as structured records to be classified, reviewed, and imported into a knowledge base.

How It Works

The workflow starts by exporting favorites to CSV, then classifies them by keywords. Low-confidence items can be rerun through an LLM for second-pass classification. In addition to the default nine categories, v1.2 supports automatic discovery of natural categories from sampled titles and accepts a custom category JSON via classify_favorites.py --categories. Incremental classification can process only new favorites, merge results, and normalize labels. Outputs include category statistics, per-category CSV files, and optional exports to an Obsidian vault, Notion database, or IMA.

Boundaries

Export and keyword classification can run offline. Automatic discovery, LLM reranking, IMA import, and Notion export require configured APIs, tokens, or credentials, and send sampled titles, summaries, or URL lists to external services. Automatic discovery works best with at least 500 favorites; smaller samples may produce incomplete categories. If the discovered taxonomy is wrong, edit user_categories.json or rerun discovery with a larger sample.