拾遗 · 通用备考错题追踪（shiyi-study-tracker）

拾遗 · 通用考试备考追踪 Skill。适用于任何考试——GRE、雅思、考研、注会、高考、期末…… 核心功能：识别错题截图 → 自由标签分类 → 词库积累复用 → 二刷提醒 → 导出 Excel。 触发关键词：做了题、错了、截图发来、导出错题、待二刷、记得、不记得、换考试。 图片消息直接触发识别。

Security Scan

The skill's code, instructions, and requested resources align with its stated purpose (a local study/tracker that recognizes screenshots, stores annotations, schedules reminders, and exports Excel); nothing indicates hidden exfiltration or unrelated credential requests.

Purpose & Capability

Name/description match the implemented files: multimodal image parsing, tag library, local JSON storage, reminders, and Excel export. The files and logic are coherent with an exam study tracker.

Instruction Scope

Runtime instructions and code operate on user-provided messages/images and local data under ~/.openclaw/skills/shiyi/data. The agentCall invocation sends image+prompt to the configured multimodal model (expected for OCR/analysis). There are no instructions to read unrelated system config or to transmit data to third‑party endpoints beyond the agentCall model invocation.

Install Mechanism

This is instruction/code-only (no platform install spec). The Node code uses a Python helper (writes and runs a temporary .py using openpyxl) when exporting Excel files but package.json does not declare Python/openpyxl dependencies. Users must have a system Python and openpyxl installed for exports to work; this is an operational requirement rather than a malicious installer.

Credentials

The skill declares no required env vars, credentials, or external config paths. It only reads/writes files under the skill data directory in the user's home directory, which is consistent with its purpose.

Persistence & Privilege

always:false and no manipulation of other skills or global agent settings. The skill persists its own data under ~/.openclaw/skills/shiyi/data (config.json, wrong_questions.json, backups, exports), which is expected for a local tracker.

Assessment

This skill appears coherent and implements the stated study-tracker features locally. Before installing, note: (1) it writes all data to ~/.openclaw/skills/shiyi/data (questions, tag library, backups, exports) — review and manage that directory if you care about privacy; (2) image recognition is performed by the agentCall/multimodal model configured in your OpenClaw environment, so screenshots and prompts are sent to whatever model endpoint your agent uses — ensure you trust that model/provider; (3) the Excel export runs a small temporary Python script that requires Python + openpyxl on your system (not declared in package.json) — install those if you need export-with-images to work; (4) the skill runs scheduled scripts (cron/job entries suggested in README) which will read/write the local data dir — verify the cron configuration and channels (e.g., Feishu) you wire it to. If any of these are concerns (remote model usage, local data retention, or additional runtime deps), review the source files or run in a sandbox before enabling for production use.

拾遗 · 通用备考追踪 Skill

一、首次安装

Skill 加载时发一条消息，问你在备考什么考试，之后不再重复。

拾遗已安装。

你在备考什么考试？比如：
  GRE、雅思、考研英语、注会、高考、期末……
  不在列表里也没关系。

输入考试名称后，Skill 加载对应的识别背景知识，之后截图识别精度更高。 如果后来换考试，发「换考试」即可重新配置，不影响已有错题记录。

---

二、与朱批录的核心差异

	朱批录	拾遗
适用范围	考公（国考/省考）	任意考试
科目结构	固定5科（言语/数量/判断/资料/申论）	自由标签，AI 自动生成
标签复用	不需要（科目是固定的）	tag_library 跨题目积累
考试切换	不支持	「换考试」指令随时切换

---

三、触发场景

用户说的话	执行
发来截图	多模态识别 + 标签归档
发来截图附带"粗心"	识别后直接归档，不追问
"Verbal-Text Completion-词汇量不足"	快捷格式，直接归档
"记得" / "不记得"	二刷自评，连续2次记得→已掌握
"导出错题本"	导出全部
"只导出待二刷的"	筛选导出
"导出Verbal的错题"	按 section 筛选
"导出最近两周的"	按时间筛选
"换考试"	重新配置考试类型

---

四、数据结构

~/.openclaw/skills/shiyi/data/
├── config.json              ← 当前考试配置
├── tag_library.json         ← 各考试的标签词库
├── wrong_questions.json     ← 所有错题
├── review_state.json        ← 二刷状态
├── review_session.json      ← 当前二刷进度
├── stats_cache.json         ← 打卡连续天数
├── daily/                   ← 每日记录
├── backups/                 ← wrong_questions 自动备份（最近10个）
└── exports/                 ← 导出文件

错题记录字段：

{
  "id": "uuid",
  "date": "2026-03-19",
  "exam": "GRE",
  "exam_name": "GRE",
  "section": "Verbal",
  "question_type": "Text Completion",
  "knowledge_point": "逻辑关系词 — 转折",
  "question_text": "完整题目文字",
  "visual_description": null,
  "answer": "B",
  "error_reason": "知识点不会",
  "keywords": ["逻辑关系词", "Text Completion"],
  "status": "待二刷"
}

---

五、核心流程

图片识别

1. 读取 config.json 里的 exam_key，加载对应的考试背景知识
2. 从 tag_library.json 取最近30天用过的标签注入 prompt（提升复用率）
3. 调用 OpenClaw 配置的多模态模型识别
4. 识别结果写入 wrong_questions.json，新标签追加进 tag_library.json
5. 调用失败 → 提示手动复制文字

标签词库（tag_library）

• 每次识别后自动追加新标签（去重）
• 同一考试内复用，不同考试互不干扰
• 每类最多保留：知识点200个、题型50个（超出删最旧的）
• 注入 prompt 时只传最近30天用过的（避免 prompt 过长）

定时推送

时间	内容
每天 21:00	当日错题汇总 + 高频考点 + 打卡天数
隔天 20:00	随机抽3道待二刷题，用户自评

---

六、文件索引

文件	作用
assets/exam_prompts.js	各考试背景知识（识别精度的核心）
scripts/tag_library.js	标签词库读写与统计
scripts/onboarding.js	首次配置考试类型
scripts/parse_input.js	消息路由 + 识别入口
scripts/update_daily.js	写入错题和每日记录
scripts/export_xlsx.js	导出 Excel（含筛选和截图嵌入）
scripts/review_reminder.js	二刷提醒和自评处理
scripts/daily_summary.js	每日定时总结

---

七、新增考试

在 assets/exam_prompts.js 的 EXAM_PROMPTS 对象里加一条字符串：

'驾照科目一': `
驾照科目一为交规理论考试，100道题，90分及格。
题型：判断题、单选题。
标 knowledge_point 时请具体到交规条款（如"禁止标志与警告标志区别"）。`,

PR 欢迎。

---

八、如果这个 Skill 对你有帮助

⭐ Star 一下，让更多备考的人找到它 🍴 Fork 加上你的考试类型，欢迎 PR

https://github.com/KaguraNanaga/shiyi