WeChat Summary Skill

把微信群聊提取做成一个可复用、可交互的摘要流程。

这个 skill 只负责:

1. 选择群聊
2. 选择时间范围
3. 选择输出模式
4. 保存最近使用的群和默认配置
5. 生成文字摘要或本地网页信息报

它不负责部署。

User Input Rule

When asking the user for choices:

1. Prefer built-in user-input tools exposed by the runtime.
2. If no such tool exists, ask concise numbered questions in plain text.
3. Batch compatible questions when possible.

Prerequisites

When running the bundled Python scripts below:

• use python3 on macOS / Linux
• use py -3 on Windows PowerShell
• use python only if that is your machine's Python 3 launcher

Before doing anything else, verify these in order:

python3 scripts/check_wechat_env.py

If the doctor fails:

1. tell the user the exact failing command
2. read references/setup-without-baoyu.md
3. guide them through upstream wx-cli install / init before continuing

This skill should reuse existing baoyu preferences when available, but it must also work without baoyu by using repo-native config.

Shared Config

This skill reads two layers of local configuration:

1. repo-native config for data_root, optional self_wxid, self_display, and optional wx_bin
2. baoyu preferences for self_wxid, self_display, and optional data_root
3. skill state for recent groups and preferred output presets

Load the skill state first:

python3 scripts/skill_state.py inspect

That command returns:

• the resolved config paths
• recent groups
• default time preset
• default summary mode
• default text/web styles
• default data root

If there is no repo-native config and no baoyu config yet, initialize one:

python3 scripts/skill_state.py init-config --scope project --data-root ./wechat

Use the returned recent_group_choices as the first question. If there are no saved groups yet, the list should effectively be Other.

Interaction Flow

1. Pick the group

Start from the saved recent groups returned by scripts/skill_state.py inspect.

• If the user picks a recent group, reuse its group_id and group_name.
• If the user picks Other, ask for the target group name or keyword, then run:

wx contacts --query "<group keyword>" --json

Filter to chatrooms only (username ending in @chatroom).

• If one match remains, use it.
• If multiple match, ask the user to choose.
• If none match, fall back to:

wx sessions --json

and search there before giving up.

The user's chosen group should be saved back into the recent-group list after the run.

2. Pick the time range

Offer these presets first:

• 1d - 1 day
• 3d - 3 days
• 7d - 1 week
• 14d - 2 weeks
• 30d - 1 month
• custom

Use the saved default preset as the recommended option.

For preset ranges, resolve absolute dates with:

python3 scripts/resolve_time_range.py --preset 7d

For custom, ask the user for since and until in YYYY-MM-DD, then run:

python3 scripts/resolve_time_range.py --since YYYY-MM-DD --until YYYY-MM-DD

3. Pick the output mode

Offer:

• text - structured text digest
• webpage - local web digest

Use the saved default summary mode as the recommended option.

4. Pick the style

Use the saved style default for the chosen mode.

Current built-in styles:

• text mode: growth-brief-v1
• webpage mode: daily-report-v1

Do not ask an extra style question unless:

• the user explicitly asks to change style, or
• more than one style is available for that mode

5. Save the session defaults

After the user confirms group + time range + mode, persist them:

python3 scripts/skill_state.py save-session \
  --scope project \
  --group-id "<group_id>" \
  --group-name "<group_name>" \
  --duration-preset 7d \
  --summary-mode text \
  --text-style growth-brief-v1 \
  --web-style daily-report-v1

Rules:

• Keep only the most recent groups near the front of the list.
• Default to project scope unless the user explicitly asks for global reuse.
• If the user asks for global reuse across projects, use xdg scope.

Fetch + Analyze

Once group and date range are fixed, always build a local analysis bundle first:

python3 scripts/prepare_wechat_digest.py \
  --chat "<group_name>" \
  --since YYYY-MM-DD \
  --until YYYY-MM-DD \
  --data-root "<data_root>"

This writes:

• raw/*.messages.json
• raw/*.stats.json
• analysis/*.analysis.json
• analysis/*.briefing.md

Read the generated briefing before opening raw messages. It is the cheapest way to understand the selected range.

Output Modes

Text mode

Read:

• references/text-summary-format.md

Use the analysis bundle plus raw messages only when needed, then write a markdown digest to:

<group_dir>/<since>_<until>.text-summary.md

The final text output must follow the section structure in references/text-summary-format.md.

Webpage mode

Read:

• references/webpage-mode.md
• references/summary-schema.md

Then:

1. Draft summary.json
2. Render the local site with:

python3 scripts/render_web_digest.py \
  --summary /abs/path/to/summary.json \
  --analysis /abs/path/to/analysis.json

This writes local output only:

• <group_dir>/<since>_<until>.web.md
• <group_dir>/site/index.html
• <group_dir>/dist/index.html
• <group_dir>/history.json

Do not add deployment steps.

When To Read Raw Messages

The analysis bundle is usually enough. Open raw messages only when:

• a quote needs verification
• a hotspot feels too vague
• you need the exact wording for demand / connector attribution
• the same nickname might refer to multiple people

Do not dump the whole raw payload into context if the date range is large. Use jq slices or targeted searches.

Output Discipline

• Use real group names and real participant names.
• Keep counts and dates accurate.
• Keep explicit facts separate from your inference.
• Do not fabricate links.
• Do not add deployment advice unless the user explicitly asks for it.