How to use the WeChat Group Chat Digest agent skill (baoyu-wechat-summary)

What the skill makes the agent do

Agent skills are folders of instructions that an AI agent loads when a request matches them. This one tells the agent how to turn scattered WeChat group messages into a readable briefing. It does not contain scripts; the actual reading of chat data is done by the external wx-cli binary (the wx command).

Between runs the skill keeps per-group history (history.json and history-digests.jsonl), per-user profiles, and per-group fact notes (memory.md), so a later digest can continue from where an earlier one stopped. A normal digest is the default. A sharper roast (毒舌) version is produced only when you ask for it.

Requirements before you install

The skill depends on several things being in place. Check them first so you do not get stuck partway through.

The wx-cli binary must be installed and working, from https://github.com/jackwener/wx-cli. The skill's quick check is that wx --version prints output and wx sessions returns data.

Commands must run with the sandbox disabled, because wx-cli reads from ~/.wx-cli/ and WeChat's local data directory. The skill itself asks the agent to run these commands with dangerouslyDisableSandbox set to true. You can view or edit sandbox restrictions with /sandbox.

On first run the skill needs a local EXTEND.md preferences file containing self_wxid and self_display. The skill tries to discover these itself and otherwise asks you to confirm them.

• wx-cli binary (the wx command) installed and working
• Access to WeChat's local data directory and ~/.wx-cli/
• Sandbox disabled for wx commands (dangerouslyDisableSandbox)
• EXTEND.md with self_wxid and self_display (created on first run)

Install the skill (ZIP or Claude apps)

Option 1: download the ZIP for this skill from this page and unzip it into your user skills folder, ~/.claude/skills/, or into a project's .claude/skills/ folder for that project only. After unzipping, the folder should contain SKILL.md and a references/ directory.

Option 2: in the Claude apps, open Settings, go to Capabilities, then Skills, and upload the skill there.

The skill's own files are SKILL.md plus references/setup.md, references/output-formats.md, references/profiles.md and references/group-memory.md. The references are loaded only when a step needs them.

Trigger the skill with natural requests

You do not need a special command. The skill activates when your request matches its description. Phrases the source lists include 总结群聊, 群聊精华, 群聊摘要, summarize group chat, group chat digest, and 帮我看看 XX 群最近聊了什么. Mentioning a group name with a time range works too.

To get the roast version, add a phrase such as 毒舌版, roast 版 or 再来个毒舌的. To backfill profiles, ask to 回溯画像 or 初始化画像 (backfill profiles).

• Summarize the 产品讨论 group from the last 7 days (replace with your group's name)
• 帮我看看 XX 群最近聊了什么
• XX 群有什么值得看的
• Continue from last time for the XX group (incremental mode)
• Do the 3 March to 5 March digest for XX 群 in the 毒舌版

What a run does

When you ask for a digest, the agent parses your time range, finds the group through wx contacts (matching usernames ending in @chatroom) or wx sessions, and then reads the messages. If the request has no time range, it continues incrementally from the last digest, falling back to default_time_range from EXTEND.md or the last 24 hours if no history exists.

The normal version is written first and includes an opening summary, categorized topic sections, an optional pain-point section, an optional @bot Q&A section, a stats block with a Top 10 leaderboard, and member portraits. Output files are written into a per-group folder under data_root, which defaults to {project_root}/wechat.

Configure preferences (EXTEND.md)

Preferences live in the first EXTEND.md found, checked in this order: .baoyu-skills/baoyu-wechat-summary/EXTEND.md in the project, then ${XDG_CONFIG_HOME:-$HOME/.config}/baoyu-skills/baoyu-wechat-summary/EXTEND.md, then $HOME/.baoyu-skills/baoyu-wechat-summary/EXTEND.md. The file uses key: value or key=value lines with # for comments.

The keys are self_wxid, self_display, default_version (normal, roast or both), default_time_range (for example 7d or 24h), data_root, and bot_aliases (default bot, 精华bot). Choose bot aliases that do not match any real group member or existing bot.

Safety and privacy notes

The skill writes chat-derived files to disk: history, digests, per-user profiles and group memory. Keep this in mind if local files holding chat content are not acceptable in your environment. The source says it has privacy guardrails built in, and profiles are kept separate for the normal and roast versions so snark does not leak into the sober summary.

The skill stores per-user profiles of other people, so only use it on groups you are a member of and where members have consented. The source states the skill makes no network calls and runs no bundled scripts; it relies on the local wx-cli binary.

• No bundled scripts; it uses the external wx binary
• Network use: none stated by the skill
• Profiles are kept separate: profiles/ for normal, profiles-roast/ for roast
• The skill never installs software or runs sudo on your behalf; it stops and gives you the exact fix instead

Tips for better digests

Name the group and a time range in your request so the agent does not have to guess. If a group name matches several groups, the agent asks you to choose. Set default_version in EXTEND.md if you usually want one style. Use incremental requests such as 接着上次 to avoid repeating what you already read.

• Give a specific group name and date range
• Set default_version once in EXTEND.md
• Use 接着上次 for daily follow-ups
• If a wx command fails, read references/setup.md for the troubleshooting steps