Open‑Source AI Digital Employees for SMBs: One‑Line Install for Claude Code/Codex

Project Repository

GitHub: https://github.com/tjxj/z-skills

The repository initially contains five Skills and will be updated continuously.

One‑Line Installation

Clone the repository into the project's skill directory and enable the Skills. After this step the Agent handles cloning, reading SKILL.md, installing dependencies and registration automatically.

git clone https://github.com/tjxj/z-skills.git skill && enable

1. z-mail-reader – Email Reading & Real‑Time Listening

Connects to any IMAP mailbox, batch‑pulls emails within a specified time range, downloads all attachments, extracts embedded images (CID, external, Base64), and outputs structured JSON for summarisation. Supports 30‑second polling for real‑time notifications and automatically handles UTF‑8, GBK, and GB2312 encodings.

One‑time configuration (example):

帮我把 z-mail-reader 需要的邮箱环境变量配上，IMAP 地址 imap.qq.com，邮箱 [email protected]，授权码 xxx

Typical commands:

读一下最近 7 天的邮件，给我生成一份摘要
拉一下这周邮件，附件都下下来
看一下 6 月 1 号到今天的邮件，重点提取合同类邮件
开始监听邮件，新邮件来了就推个系统通知给我
停一下邮件监听

2. z-smart-xparse – Intelligent Document Parsing

Automatically splits large PDFs (tens of megabytes, hundreds of pages) into 50‑page chunks (adjustable to 200 pages with paid APIs), parses each chunk, and merges results. Supports PDF, images, Word, PPT, Excel, HTML, OFD, RTF, with optional OCR for scanned PDFs.

Decision logic:

If size ≤ 5 MB and pages ≤ 100 → parse directly.

If size > 5 MB or pages > 100 → split → parse each chunk → merge.

Typical one‑line commands:

把这份 report.pdf 转成 markdown
这份 300 页的招标书读一下，提取关键资质要求和评分规则
只解析 contract.pdf 的前 5 页
这是张扫描件合同，帮我转文本并提取表格

3. z-md-excel – Markdown Table → Excel

Detects all GFM‑style pipe tables in a markdown file, creates a separate sheet for each (named Table_1, Table_2, …), preserves alignment, strips markdown formatting, and outputs a professionally formatted Excel file with blue header styling, frozen top row, and auto‑sized columns. Only GFM pipe tables are supported; HTML tables and nested tables are ignored.

Typical commands:

把 xx.md 里的表格导出成 Excel
这份调研笔记里有三个对比表，都给我转成 Excel
把上面生成的 Markdown 表格保存为《产品调研.xlsx》

4. z-excel-editor – Full‑Featured Excel Editing

Built on openpyxl and LibreOffice. Capabilities include creating new workbooks, editing existing files, adding formulas, applying financial‑model colour conventions (blue = input, black = formula, green = cross‑sheet reference, red = key assumption), enforcing numeric formatting, and automatically detecting and reporting formula errors ( #REF!, #DIV/0!, #VALUE!, #NAME?, #N/A) with JSON error locations.

Typical commands:

把这份 sales.xlsx 加上表头、合计公式和环比增长率
这份财务模型帮我扫一下公式错误，能修的顺手修了
把这张估值表按金融模型颜色规范重新整理（输入值蓝色、公式黑色、跨表引用绿色）
新建一份 2026 年运营预算表，带季度合计和全年合计公式

5. z-web-pack – Web Asset Collection Pack

Collects links and images from a set of URLs, preserving hierarchy and storing assets locally. Parameters --title, --max-depth, and --max-pages control the crawl; unreachable pages fall back to jina.ai extraction.

Typical command:

把这些链接采集为《竞品分析》素材包：xxx

Resulting directory structure:

2026-06-08-竞品分析/
├── README.md            # 素材包概览
├── 00-research-brief.md
├── 01-link-inventory.md
├── 02-image-inventory.md
├── 03-reading-map.md
├── MAIN-01-入口正文.md
├── LINKED-02-相关链接.md
└── assets/              # 本地图片

Customization Requirement

The Skills are not universal plug‑and‑play templates; they must be adapted to each company's email system, document formats, and naming conventions. For example, the demo uses QQ mail (IMAP imap.qq.com); enterprises may need to replace the IMAP address, adjust OCR settings, or modify output paths.

Chaining Skills into a Full Workflow

When linked, a single natural‑language request triggers the entire pipeline: email retrieval → PDF parsing → markdown‑to‑Excel conversion → Excel polishing. The Agent handles parameter passing and orchestration.

Illustrative chain:

邮件到了 → z-mail-reader 自动拉取
    ↓
附件是 PDF → z-smart-xparse 解析成 Markdown
    ↓
Markdown 里有表格 → z-md-excel 导出 Excel
    ↓
Excel 需要加工 → z-excel-editor 添加公式、美化格式
    ↓
最终产出一份可直接汇报的 Excel 文件

Three concrete scenarios:

Competitor weekly report: z-web-pack → z-md-excel → z-excel-editor.

Bid document compliance: z-mail-reader → z-smart-xparse → Agent‑generated compliance report.

Financial month‑end close: z-mail-reader → z-smart-xparse → z-excel-editor.

Future Integration

Planned extensions include pushing email summaries to Feishu groups, writing parsed data into Feishu multi‑dimensional tables, and scheduled competitor data collection that updates Feishu documents.

Conclusion

The open‑source Skill set converts AI Agents from chat interfaces into reproducible, auditable digital employees that execute repeatable processes across multiple Agent platforms, offering reproducibility, debuggability, vendor‑agnosticism, and composability.