larkin0302/group-daily

group-daily is a Claude Code Skill that turns a day, week, or month of chat in a WeChat group into a magazine-style report rendered as an HTML page and a long PNG screenshot you can post elsewhere. The README is in Chinese. The author's pitch is that existing AI summarizers for WeChat groups produce dry meeting-minute bullet lists nobody wants to share, while products like Spotify Wrapped show that a narrative format with a timeline, highlighted characters, and a visual style gets passed around. This skill applies that idea to chat logs. The most sensitive part is how it reads WeChat data. The desktop WeChat app on macOS stores messages, contacts, and avatars in local SQLite files that are encrypted with SQLCipher, so you cannot just open them. The skill expects the files to already be decrypted by a separate tool. The author has a private project called wechat-decrypt that bundles a CLI named vchat and an MCP server, and the skill prefers that path. If you do not have it, the README points to open-source alternatives like PyWxDump and explains how to drop their decrypted output into the folder layout the skill expects, using a VCHAT_DATA_DIR environment variable. Installation is clone the repo into ~/.claude/skills/group-daily and run install.sh, which installs Python packages including Pillow, openai-whisper, and silk-python, and runs a six-item environment self-check. You then ask Claude Code in plain language to make a report for a given group name. The pipeline has eight steps: pull the chat log, transcribe voice messages locally with Whisper, compute basic stats, load any saved style fingerprints for the group, write a story.json with a six to eight beat timeline plus highlighted speakers and a Q and A section, look up group member IDs, render the HTML, and convert it to a PNG using headless Chrome. Nothing is uploaded; everything runs on your machine.