WeRead Reading Recommender#

Overview#

This is a local-first skill for exporting 微信读书 (WeRead) records from a cookie stored on the user's machine, normalizing those records into a recommendation-friendly JSON file, and using that data to analyze reading preferences or recommend what to read next.

Use this skill when the user wants to:

• 根据微信读书记录推荐书
• 分析自己的阅读偏好或阅读画像
• 结合“最近想学的主题”与微信读书历史一起做推荐
• 导出、刷新、归一化本地微信读书数据

Trigger Cases#

Activate this skill for requests like:

• “根据我的微信读书记录推荐书”
• “分析我的阅读偏好”
• “我最近想系统学 AI Agent，结合微信读书记录推荐 5 本书”
• “帮我导出 / 刷新 / 归一化微信读书数据”
• “基于我的阅读历史，推荐下一本最适合现在读的书”
• “分析我的阅读偏好，并给我 3 本稳妥推荐 + 2 本探索推荐”

Workflow#

Follow this sequence:

1. Check whether a normalized JSON file already exists.
2. If normalized data is missing, or the user explicitly wants fresh data, check whether a local WeRead cookie is already available.
3. Look for a local cookie source in this order:
   • a cookie file path explicitly provided by the user
   • WEREAD_COOKIE
   • another env var name passed through --env-var
4. If no local cookie source exists, ask the user to set one locally and stop there. Do not tell the user to edit SKILL.md.
5. If a local cookie source exists, run the export script.
6. Run the normalize script on the raw export.
7. Read the normalized JSON and identify strong signals:
   • high-engagement books
   • recent books
   • unfinished books with momentum
   • repeated categories or lists
8. If the user provides a current goal, weight goal fit first.
9. If the user does not provide a goal, produce a reading-profile summary plus safe and exploratory recommendations.

Recommendation Guidance#

When the user provides a current goal, weight approximately:

• 60% goal fit
• 40% history fit

When the user provides no goal, weight approximately:

• 70% history fit
• 20% recency
• 10% exploration/diversity

For each recommendation, explain:

• why it fits the user's current goal or history
• which past books it resembles
• what gap it fills
• whether it is a safe pick or an exploration pick
• whether it is a good fit right now

Suggested response structure:

• 阅读画像 / Reading profile
• 推荐结果 / Recommendations
• 为什么适合现在 / Why now
• 暂缓推荐 / Skip for now (optional)

Local Data Workflow#

1. Check local cookie availability first#

Before asking the user to set anything, first check whether a local cookie is already available through:

• a cookie file path the user provided
• WEREAD_COOKIE
• another env var name passed through --env-var

If none of these exist, ask the user to set the cookie locally, then continue.

2. Export raw WeRead data#

If a local cookie is already available, export directly:

bash

python3 scripts/export_weread.py --out data/weread-raw.json

Optional variants:

bash

python3 scripts/export_weread.py --cookie-file ~/.config/weread.cookie --out data/weread-raw.json
python3 scripts/export_weread.py --env-var WEREAD_COOKIE --include-book-info --detail-limit 50 --out data/weread-raw.json

If the user does need to set one manually, keep it local. For example:

bash

export WEREAD_COOKIE='wr_skey=...; wr_vid=...; ...'

3. Normalize the raw export#

bash

python3 scripts/normalize_weread.py --input data/weread-raw.json --output data/weread-normalized.json

4. Use the normalized file for recommendation turns#

After normalization, this skill should reason primarily from the normalized JSON, not from a live cookie session, unless the user explicitly asks for a refresh.

Security Boundary#

This skill is local-first. Enforce these rules:

• Cookie is for local use only.
• Never write the cookie into SKILL.md, scripts, assets, logs, or exported JSON.
• Never echo the cookie back in responses.
• Prefer checking existing local cookie sources before asking the user to set one again.
• Do not rely on CookieCloud or any third-party cookie sync service by default.
• Do not suggest remote cookie hosting as the normal path.
• Recommendation work should use the normalized JSON whenever possible.

Files#

Use these project files as the main references:

• scripts/export_weread.py
• scripts/normalize_weread.py
• references/data-schema.md
• references/privacy-model.md
• references/recommendation-rubric.md
• assets/sample-weread-raw.json
• assets/sample-weread-normalized.json

Example Requests#

• 结合我的微信读书记录，我最近想系统学 AI Agent，推荐 5 本书
• 基于我的阅读历史，推荐下一本最适合现在读的书
• 分析我的阅读偏好，并给我 3 本稳妥推荐 + 2 本探索推荐
• 帮我刷新微信读书数据，然后按最近在读主题推荐下一批书