Crypto Learning

加密货币自学系统。每天早上9点自动推送学习内容；每次调用都必须产出小白友好、可直接学习的详细报告（不是只给标题），并通过 web_search 检索并整合最新资料。包含完整学习大纲（小白向、投资向、进阶投资三个阶段），支持进度跟踪、跳过与重置。

Security Scan

The skill promises daily, web_search-backed, long structured learning reports and strict output rules, but the bundled code does not perform web searches or generate the required detailed reports—therefore behavior and claims are inconsistent.

Purpose & Capability

The manifest and SKILL.md claim the agent must perform web_search, integrate latest sources, and always produce long (>=450字) multi-section lessons. The bundled scripts (crypto_learning.py and learn.py) only read local content.json/progress.json and output short, simple messages or JSON objects. There is no implementation of web_search, network retrieval, nor enforcement of the detailed output template and hard constraints. The README/schedule.sh mention sending via a message tool, but no credentials or real sending implementation exist. This mismatch indicates the skill will not deliver the advertised capabilities without additional platform features or code changes.

Instruction Scope

SKILL.md instructs reading progress.json/content.json (which the code does) and explicitly requires web_search and synthesizing multiple external sources. No code calls network or browser APIs; schedule.sh only runs local scripts and writes to /tmp. The SKILL.md also imposes strict output formatting rules (6 modules, term explanations, 3 self-test questions, >=450字), none of which are validated or produced by the included code. If the agent runtime provides a separate web_search capability and is expected to follow SKILL.md instructions at call time, that could satisfy some behavior, but the packaged code alone does not implement or enforce the described scope.

Install Mechanism

No install specification is provided (instruction-only plus scripts). No external download, package installation, or archive extraction is present in the bundle. This minimizes install-time risk—nothing external is automatically fetched by an installer.

Credentials

The skill declares no required environment variables, no credentials, and no config paths. The code references only local files inside the skill (content.json, progress.json) and writes a temporary file /tmp/crypto-learning-today.txt in schedule.sh. There is no apparent request for unrelated secrets or privileged credentials.

Persistence & Privilege

always is false and the skill does not request permanent platform-level privileges. It persists progress to progress.json within the skill directory (expected behavior for a learning tracker). schedule.sh suggests a cron-driven push but does not actually register a system-wide cron job—deployment/configuration would be required. No modification of other skills or system-wide configs is present.

What to consider before installing

This skill's description promises automated daily lessons that fetch and synthesize web sources and always produce long, structured reports, but the included scripts only read local content/progress and print short messages. Before installing or relying on it: - Decide which behavior you expect: if you need live web searches and the strict report format, you'll need either (A) the platform's web_search/browser capability hooked into the agent at runtime and the agent instructed to follow SKILL.md exactly, or (B) the skill code modified to perform web requests and generate the long formatted reports. The bundle as-is will not perform web searches. - Test locally: run learn.py and crypto_learning.py commands (today/next/status/reset) to see the actual outputs and how progress.json is updated. - Messaging/scheduling: schedule.sh just writes to /tmp and echoes content; it does not actually send messages. If you expect automatic 9:00 pushes to Telegram or another channel, implement and secure the message-sending step and set up a proper scheduler. - Data persistence: progress.json is stored in the skill directory and will be modified by the scripts—backup if you care about the data. - Security posture: no credentials are requested and no external installers are used, so direct exfiltration risk from this bundle is low; however, if you or a maintainer extend the skill to call external services or add message-sending (e.g., Telegram API), inspect those additions and protect any tokens. If you want this to behave as advertised, ask the author to (1) implement the web_search/synthesis step or explicitly state that the platform's web_search will be used at runtime, and (2) implement/enforce the SKILL.md output constraints or remove them to match actual code.

Crypto Learning - 加密货币自学系统

目标

给 hmzo 提供“每天可直接学习”的加密货币课程推送：

• 解释清楚（小白能看懂）
• 内容完整（不是只有题目）
• 有最新资料整合（web_search 检索）
• 有可执行学习动作（自测/作业）

学习大纲位置

学习大纲在 content.json：

• 小白向（30天）
• 投资向（60天）
• 进阶投资（90天）

进度在 progress.json。

必须遵守的输出质量规则（硬约束）

每次输出学习报告都必须满足以下要求：

1. 禁止只输出题目或一句话摘要。
2. 最少包含 6 个模块（见下方固定模板）。
3. 正文长度不少于 450 字（不含标题与分隔符）。
4. 术语必须解释：首次出现术语（如 PoW、PoS、Gas、TVL）时，给一句“白话解释”。
5. 至少给 2 个生活类比或场景例子，帮助理解抽象概念。
6. 至少给 3 条要点总结（可背诵版本）。
7. 必须含 3 道自测题 + 标准答案。
8. 最后给出次日预告，并更新学习进度。

若 web_search 检索结果不足，仍要基于 content.json 产出完整报告，不得降级为短消息。

固定输出模板（每次都用）

按以下结构输出（允许微调措辞，但模块不可缺失）：

1. 📚 第 X 天学习内容（阶段 / 主题 / 今日重点）
2. 📖 先讲人话版（先用 3-6 句白话解释核心概念）
3. 🧱 系统化知识（定义、原理、关键组成、常见误区）
4. 🌐 最新资料整合（基于 web_search 的信息融合，不要贴生硬链接堆砌）
5. 🧠 例子与类比（至少 2 个）
6. ✅ 今日小结（至少 3 条可记忆要点）
7. 📝 3题自测（含答案）
8. 🚀 明日预告 + 进度更新

工作流程

1) 读取进度

读取 progress.json：

• current_stage
• current_topic_index
• current_subtopic_index
• completed_subtopics

2) 获取今日主题

根据进度从 content.json 取出：

• 阶段
• 主题
• 子主题
• 基础知识点

3) web_search 检索并整合

必须检索当日主题相关的最新内容，再进行融合输出。

检索建议：

• 中文关键词：子主题 + 加密货币 + 入门/教程/原理
• 英文关键词：subtopic + crypto + beginner guide/explained
• 优先权威来源：Binance Academy、CoinMarketCap Alexandria、Investopedia、Ethereum/Bitcoin 官方文档等

要求：

• 不照搬原文；用自己的结构重写
• 把多来源信息合并成统一叙事
• 说明“共识点”和“容易误解点”

4) 生成详细学习报告

严格使用“固定输出模板”，并满足“硬约束”。

5) 更新进度

完成后更新 progress.json：

• 标记当前子主题完成
• 指向下一个子主题
• 更新 last_push_date
• total_days_completed +1

用户命令

开始学习计划

按当前进度继续推送。

查看进度

显示：已完成天数、当前阶段/主题、已完成知识点。

跳过今天

记录跳过日期，不推进学习内容。

重置计划

清空进度并从第一个主题开始。

手动学习某个主题

支持临时学习指定主题；默认不改自动进度（除非用户明确要求）。

失败降级策略

• web_search 失败：改用 content.json + 已有常识生成完整详细版，不得只发标题。
• 信息冲突：优先通用共识，标注“不同平台口径可能不同”。
• 内容过深：自动加“白话翻译”段，确保小白可读。

风格要求

• 默认中文，语气亲切但不废话。
• 先白话后术语，先结论后展开。
• 以“能学会”为目标，不写成新闻快讯。
• 每次都要让用户“看完就有收获”。