← Back to Skills Marketplace
kagurananaga

考公备考追踪

by KaguraNanaga · GitHub ↗ · v1.0.3 · MIT-0
cross-platform ⚠ suspicious
271
Downloads
0
Stars
0
Active Installs
4
Versions
Install in OpenClaw
/install kaogong-study-tracker
Description
朱批录 · 国考备考追踪 Skill。当用户发来套题成绩、错题截图、备考打卡或复习进度时触发。 核心功能:识别错题截图 → 分类错题原因 → 更新本地记录 → 生成每日总结 → 导出 Excel / 同步飞书。 触发关键词:做了一套题、今天做了、错了几道、帮我分析、备考打卡、行测、申论、 判断推理、资料分析、言语...
README (SKILL.md)

朱批录 · 国考备考追踪 Skill

一、首次安装提示

Skill 首次加载时(~/.openclaw/skills/kaogong-study-tracker/.welcomed 不存在), 主动发一条说明消息,之后不再重复:

朱批录已安装。

直接发文字就能记录,比如"今天判断推理错了8道"。
发截图的话,需要 OpenClaw 配置了支持图片输入的多模态模型才能自动识别。
没有的话也没关系,把题目文字手动复制过来发给我,一样能整理。

不问任何问题,不存储任何凭据。


二、概览

平台无关——飞书、Telegram、WhatsApp、Discord,逻辑完全一致。

图片识别统一走多模态模型:文字题、图形推理、统计图表都能理解,不依赖本地 OCR。


三、触发场景

用户说的话(示例) 应执行的操作
"今天做了一套行测,判断推理错了8道" → 解析 + 归档 + 分析
发来一张错题截图(图片消息) → 多模态识别 + 单题归档 + 追问原因
发来截图并附带"粗心" → 多模态识别 + 直接归档,不追问
"把今天的错题发给你:第12题……" → 错题分类 + 存档
"今天申论没写,太累了" → 打卡记录(未完成状态)
"我最近资料分析一直不稳,怎么办" → 查历史记录 + 建议
"帮我看看最近哪个模块最弱" → 统计分析 + 回复
"导出错题本" / "把错题发给我" / "生成报告" → export_xlsx.js,发回文件
"只导出待二刷的" → 筛选导出,仅待二刷题目
"导出判断推理的错题" → 按科目筛选导出
"导出最近两周的" → 按时间筛选导出
"只导出待二刷的资料分析题" → 多条件组合筛选导出
"资料-乘积增长-公式不熟-待二刷"(快捷格式) → 直接归档,不追问
二刷时回复"记得" / "不记得" → review_reminder.js 处理,连续2次记得→已掌握
"同步到飞书" / "更新飞书错题本" → feishu_doc.js,同步含截图

四、数据结构

所有数据以 JSON 存储在 ~/.openclaw/skills/kaogong-study-tracker/data/

4.1 每日记录 daily/{YYYY-MM-DD}.json

{
  "date": "2026-03-17",
  "modules": {
    "言语理解": { "wrong": 6,  "total": 40 },
    "数量关系": { "wrong": 8,  "total": 15 },
    "判断推理": { "wrong": 10, "total": 40 },
    "资料分析": { "wrong": 7,  "total": 20 },
    "申论":     { "written": false }
  },
  "mood": "中性",
  "note": "用户原话"
}

4.2 错题本 wrong_questions.json

[
  {
    "id": "uuid",
    "date": "2026-03-17",
    "source": "image",
    "module": "判断推理",
    "subtype": "逻辑判断",
    "question_text": "题目文字;图形题写对规律的描述",
    "visual_description": "图形推理/统计图的详细视觉描述(多模态模型生成)",
    "answer": "B",
    "user_annotation": "用户手写批注",
    "error_reason": "知识点不会 | 粗心 | 时间不够 | 概念混淆",
    "keywords": ["假言命题", "逆否命题"],
    "raw_image_b64": "base64...",
    "status": "待二刷 | 已掌握"
  }
]

4.3 统计缓存 stats_cache.json

{
  "last_updated": "2026-03-17",
  "streak": 5,
  "total_days_studied": 12,
  "weak_modules": ["数量关系", "判断推理"],
  "module_accuracy": {
    "言语理解": 0.82,
    "数量关系": 0.51,
    "判断推理": 0.68,
    "资料分析": 0.74
  }
}

五、核心流程

Step 1:消息路由(parse_input.js

文字消息 → parseStudyInput()   提取科目/错题数/情绪
图片消息 → parseImageInput()   调用多模态模型

图片处理流程:

  1. 读取 config.json 中的 multimodal 配置
  2. 如未配置 → 回复"请先配置多模态模型 API,见安装提示"
  3. 调用多模态模型,提取:科目、题型、题目内容、视觉描述、答案、错误原因推测
  4. needs_confirm 不为 null 时追问(最多一个问题);caption 已含原因则直接归档

追问只问一次,按优先级:

  • 识别不到科目 → 问科目
  • 原因不明确 → 问"粗心还是没掌握还是时间不够"
  • 信息完整 → 不追问,直接归档

Step 2:归类错题原因

原因 关键词
知识点不会 不懂、没学过、概念不清楚
粗心 看错、算错、选反了
时间不够 没做完、最后几题蒙的
概念混淆 搞混了、分不清、以为是

Step 3:更新记录(update_daily.js

写入 daily/{date}.json,同步更新 stats_cache.json(连续打卡、模块准确率)。

Step 4:生成回复

references/reply_templates.md,150 字以内。语气见 references/tone_guide.md

Step 5:导出 Excel(export_xlsx.js

  • 截图原图通过 openpyxl(Python)嵌入对应行
  • Windows 兼容:先尝试 python3,失败自动 fallback 到 python
  • 输出两个 Sheet:错题本(含截图列)+ 每日记录
  • 可直接发给 Kimi / 其他模型做趋势分析

Step 6:同步飞书云文档(feishu_doc.js,可选)

  • 需配置 feishu_doc.app_id / app_secret / doc_token
  • 截图上传飞书文件系统后作为图片块插入文档
  • 对图形推理、统计图最有用:飞书内直接看图,不用下载

Step 7(可选):定时推送

每天 21:00 触发 daily_summary.js,自动发当日总结。


六、文件索引

文件 作用
scripts/parse_input.js 文字解析 + 多模态图片识别
scripts/update_daily.js 写入每日记录 + 统计缓存
scripts/export_xlsx.js 导出 Excel(含截图嵌入,openpyxl)
scripts/feishu_doc.js 同步到飞书云文档(含图片块,可选)
scripts/daily_summary.js 定时汇总并主动发送
references/reply_templates.md 回复话术模板
references/tone_guide.md 语气风格指引
assets/module_map.json 科目/模块名称标准化映射
assets/config.example.json 配置模板(多模态 + 飞书),复制为 config.json 使用

七、错误处理

  • 未配置多模态 API 却发图片 → 回复安装提示,引导配置
  • 模型识别返回 error → 回复"没识别出来,能文字描述一下题目吗?"
  • 文字消息解析不出科目 → 回复"能说说今天做了哪个科目、错了几道吗?"
  • 数据写入失败 → 记录 error log,回复"记录暂时存不上,你提醒我稍后再试"
  • 连续 3 天无打卡 → 下次收到消息时,回复末尾轻轻提一句

八、隐私说明

所有数据(含截图 base64)存储在本地,不上传任何云端(飞书同步除外,仅在用户主动触发时上传到用户自己的飞书文档)。


九、如果这个 Skill 对你有帮助

⭐ Star 这个仓库,让更多备考的人能找到它 🍴 Fork 改成你的考试类型(省考 / 事业单位 / 军考……) 有问题欢迎提 Issue 或 PR。

https://github.com/KaguraNanaga/kaogong-study-tracker

Usage Guidance
This skill appears to do exactly what it says: it stores your study logs and screenshots locally, can export an Excel with embedded screenshots, and can optionally upload selected data to your Feishu account if you configure Feishu app credentials. Before installing or enabling optional features, consider: - Privacy: screenshots are stored as base64 in ~/.openclaw/skills/kaogong-study-tracker/data/ (wrong_questions.json). If you keep sensitive content in images, do not enable cloud sync. - Feishu sync: only occurs if you create/configure feishu settings in config.json; it will upload images to your Feishu app/account. Only provide app_id/app_secret/doc_token if you trust that destination. - Multimodal model use: to auto-parse images the skill sends images/text to whichever multimodal model you configure (remote models will receive the image-derived data). If you prefer fully local processing, configure/use the included OCR (scripts/ocr_image.py with PaddleOCR) but note that installing paddlepaddle/downloaded models can be large. - Optional dependencies: exporting with embedded images runs a dynamically generated Python script and requires openpyxl (the script assumes you have a working python/python3 and openpyxl). The skill’s package.json provides pip/npm hints; these installs are normal but can be heavy (PaddlePaddle). - Minor docs/code mismatch: docs say “doesn't rely on local OCR”, but a local OCR fallback (ocr_image.py) is present — behavior depends on your configuration. If you want to proceed: inspect and control config.json before enabling Feishu/multimodal models, and be comfortable with local storage of image data. If you do not want any cloud transmission, avoid configuring remote multimodal model API keys and Feishu credentials.
Capability Analysis
Type: OpenClaw Skill Name: kaogong-study-tracker Version: 1.0.3 The 'kaogong-study-tracker' skill is a legitimate tool designed to help users track civil service exam preparation. It features local data persistence, automated study summaries via cron jobs, and utilities for exporting records to Excel or syncing with Feishu (Lark). While the skill performs network requests (to Feishu APIs) and executes a dynamically generated Python script (for Excel formatting in `scripts/export_xlsx.js`), these actions are transparently documented, require user-provided credentials, and use proper serialization (JSON.stringify) to prevent injection. No evidence of data exfiltration, malicious prompt injection, or unauthorized access was found.
Capability Assessment
Purpose & Capability
The name/description (exam study tracker: capture wrong-question screenshots, classify reasons, update local records, export Excel, optionally sync to Feishu) matches the included scripts and data layout. Required capabilities (local JSON storage, image handling, optional Feishu sync and optional multimodal model calls) are consistent with the stated purpose.
Instruction Scope
Runtime instructions and code operate within scope: parse messages, optionally call a multimodal model, run local OCR fallback, persist JSON under ~/.openclaw/skills/.../data, generate exports and optionally upload to Feishu. One minor inconsistency: SKILL.md/README emphasize 'not depending on local OCR', yet a local OCR script (scripts/ocr_image.py) is included and can be invoked as a fallback. The MULTIMODAL_PROMPT forces detailed outputs (full visual_description/question_text), which means large amounts of image-derived content will be sent to whichever multimodal model you configure — this is expected for the feature but relevant to privacy.
Install Mechanism
No external arbitrary download/install spec in the skill manifest (instruction-only install). package.json lists normal JS dependencies (xlsx) and optional pip instructions for PaddleOCR. The code runs local scripts and dynamically generates a Python script for openpyxl. The only significant installs are optional: paddlepaddle/paddleocr (large model downloads) and optional sharp. No use of obscure download URLs or extracted arbitrary archives was found.
Credentials
The skill does not require environment variables by default. Optional sensitive credentials are stored in a config.json (not env vars): multimodal model API config and Feishu credentials (app_id/app_secret/doc_token) — these are proportional to optional features (remote multimodal model usage and Feishu sync). Users should be aware that enabling these features will cause images/text to be transmitted to those third parties. All image data and base64 strings are stored locally by default.
Persistence & Privilege
The skill persists data in ~/.openclaw/skills/kaogong-study-tracker/data/ (daily records, wrong_questions.json, backups, review state). It creates a local flag file on first run to avoid repeating onboarding. Croned scripts (daily_summary, review_reminder) are present but run only if enabled in workspace.yaml — the skill itself does not force always:true. This local persistence is consistent with its purpose.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install kaogong-study-tracker
  3. After installation, invoke the skill by name or use /kaogong-study-tracker
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.3
kaogong-study-tracker v1.0.3 - 更名为「朱批录」并简化首次安装流程,无需配置凭据,首次加载仅提示一次使用说明。 - 新增 scripts/review_reminder.js,支持二刷答题时根据“记得/不记得”自动更新错题状态。 - 导出支持:可按科目、时间区间、错题状态多条件组合筛选导出错题本。 - 移除 assets/config.example.json,不再需要此配置模板。
v1.0.2
kaogong-study-tracker 1.0.2 - Reduce Unnecessery questions. - Improve export xlsx functions.
v1.0.1
- 添加「安全说明 / Security Notes」段落,明确说明 API Key 管理、数据流向和隐私边界。 - 说明仅将截图发送到用户指定的多模态模型 API,不向其他地址传输数据。 - 详细列出 Skill 执行的所有本地操作,包括执行 Python 子进程、写入本地文件等。 - 其余功能描述、触发场景与数据结构保持不变。 EN Version: - Added a "Security Notes" section to clearly outline API key management, data flow, and privacy boundaries. - Clarified that screenshots are transmitted solely to the multimodal model API specified by the user, and that no data is transmitted to any other destinations. - Provided a detailed list of all local operations performed by the Skill, including the execution of Python subprocesses, writing to local files, and similar actions. - All other functional descriptions, trigger scenarios, and data structures remain unchanged.
v1.0.0
kaogong-study-tracker 1.0.0 初始发布 - 提供考公备考错题/进度追踪与自动归档,支持图片与文字多模态识别,不依赖本地 OCR。 - 首次安装自动引导配置多模态模型,按用户输入联网校验,仅需一次手动确认。 - 提供清晰的数据结构设计,所有数据本地存储,包含每日记录、错题本与统计缓存。 - 支持图片错题的智能识别、归类、追问与每日总结自动生成,并可导出 Excel、同步飞书文档。 - 跨平台适用,触发场景丰富,支持定时推送和贴心错误处理提醒。 - 坚持隐私优先,所有敏感数据本地保存,除用户主动同步外不上传云端。 v0.1.0 — 初始发布 / Initial release 首个公开版本,包含错题识别、每日打卡、错题本导出、飞书云文档同步等核心功能。 First public release. Includes wrong-answer recognition, daily check-in tracking, Excel export with embedded screenshots, and optional Feishu cloud doc sync.
Metadata
Slug kaogong-study-tracker
Version 1.0.3
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 4
Frequently Asked Questions

What is 考公备考追踪?

朱批录 · 国考备考追踪 Skill。当用户发来套题成绩、错题截图、备考打卡或复习进度时触发。 核心功能:识别错题截图 → 分类错题原因 → 更新本地记录 → 生成每日总结 → 导出 Excel / 同步飞书。 触发关键词:做了一套题、今天做了、错了几道、帮我分析、备考打卡、行测、申论、 判断推理、资料分析、言语... It is an AI Agent Skill for Claude Code / OpenClaw, with 271 downloads so far.

How do I install 考公备考追踪?

Run "/install kaogong-study-tracker" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is 考公备考追踪 free?

Yes, 考公备考追踪 is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does 考公备考追踪 support?

考公备考追踪 is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created 考公备考追踪?

It is built and maintained by KaguraNanaga (@kagurananaga); the current version is v1.0.3.

💬 Comments