✓
Purpose & Capability
名字与描述（基于职业标准构建知识图谱与学习路径）与包含的脚本和工作流一致：脚本用于提取PDF文本、构建KG、从国内权威/参考站点验证内容（如百度百科、国家站点、慕课平台）。没有请求与目的不符的凭证或外部服务。
ℹ
Instruction Scope
SKILL.md 明确描述了运行流程（提取PDF→解析章节→增量写入/tmp→多格式导出→来源验证并发送会话进度）。这与代码一致。但 SKILL.md 明确要求使用 `pdftotext` 提取全文，而 registry metadata 列出的“必需二进制/环境变量”为空——这是一个不一致（运行时可能失败）。此外，工作流程没有硬性超时且会进行网络检索/抓取（百度百科、政府/慕课站点），会产生外部网络流量并可能耗时较长；SKILL.md 要求通过“当前会话发送进度通知”，意味着运行时会反复输出中间结果。
ℹ
Install Mechanism
包是 instruction-only（没有安装规范），但包含多个 Python 脚本（scripts/）。没有声明需要的运行时（如 Python 版本、PyYAML 等第三方库）或系统工具（pdftotext）。脚本大部分使用标准库 urllib/json/re/re，但部分脚本导入了 yaml（需要 PyYAML）。没有从不可信 URL 下载依赖或执行远程二进制，这降低了直接供应链风险，但缺乏明确依赖声明会导致运行失败或意外行为。
✓
Credentials
技能未要求任何环境变量、凭证或配置路径，这与其以公开权威资源为主的设计一致。脚本会进行公开网站抓取（baike、gov、mooc），但不要求敏感凭据或访问其它技能的配置。
✓
Persistence & Privilege
flags 中没有 always:true，也未声明要修改其他技能或系统范围配置。脚本写入 /tmp 和输出目录文件，但这属于本技能自身的正常产物，未表现出对系统持久化或跨技能权限的请求。
What to consider before installing
Things to consider before installing and running:
- Missing runtime/dependency declarations: SKILL.md and scripts expect tools/libraries that are not declared in metadata. Ensure your environment provides: a suitable Python 3 runtime, PyYAML (yaml), and the pdftotext binary (poppler-utils) if you plan to process PDFs. Without these the skill may fail or behave unexpectedly.
- Network access and scraping: the code actively fetches content from public Chinese sites (baike.baidu.com, osta.org.cn, openstd.samr.gov.cn, icourse163.org, etc.). Confirm you allow outbound HTTP(S) from the agent environment and consider rate limits / terms of service for those sites.
- Long-running, no hard timeout: the SKILL.md states “no hard timeout” and incremental file writes and progress notifications; run it in a controlled/sandboxed environment to avoid runaway jobs or unexpected resource use.
- File writes: the skill writes temporary files to /tmp and produces outputs in an output_dir; review these outputs and ensure sensitive documents are not unintentionally uploaded.
- Legality/ToS: automated scraping of sites (e.g., Baidu Baike) may conflict with site terms of service—review before wide use.
- Audit the scripts: although I saw no hidden external command-and-control endpoints, you should review scripts (especially truncated/omitted files) for any hardcoded or obfuscated endpoints before granting broader permissions.
- If you need stricter safety: run the skill in an isolated container/VM with limited network access, install only required dependencies, and review logs/output files. If you want to prevent autonomous runs, keep model invocation controlled (disable autonomous invocation in your agent policy).
