Tencent Docs Reader

Read Tencent Docs spreadsheet via agent-browser copy-paste method. Supports reading any sheet tab and returns tab-separated text. 腾讯文档在线表格读取工具，支持指定子表，返回制表符分隔...

Security Scan

The skill's core reader behavior matches its description, but included helper script(s) contain hardcoded URLs and undeclared environment/webhook usage (external notifications), which is inconsistent and warrants caution.

Purpose & Capability

The main SKILL.md and read_sheet.py implement a general-purpose Tencent Docs reader via agent-browser, which is coherent. However, scripts/check_weekly_report.py is a standalone utility that hardcodes a specific DOC_URL and attempts to notify via QQ/Webhook — behavior that is not implied by the skill metadata. A generic reader should not embed a specific document URL or unattended notification workflow.

Instruction Scope

SKILL.md instructs installing and using agent-browser and running the reader script (expected). But the repo includes check_weekly_report.py which runs read_sheet.py against a hardcoded document and then may send notifications (prints QQ messages and can POST to a WECOM webhook). That expands scope from 'read a URL I provide' to 'periodically check and notify about a particular doc', which is not documented in the skill metadata and could run autonomously if invoked.

Install Mechanism

No packaged install spec is included (instruction-only), which minimizes remote-code install risk. The SKILL.md asks users to install agent-browser via npm -g, a public package; no downloads from unknown URLs or archives are present in the skill itself.

Credentials

Registry metadata declares no required environment variables, but check_weekly_report.py reads QQ_USER_ID and WECOM_WEBHOOK_URL from the environment. These are used for notifications (sending external HTTP POSTs). The presence of an undocumented default QQ_USER_ID and optional webhook is disproportionate to a document-reading skill and could result in unexpected outbound notifications if the webhook env var is set.

Persistence & Privilege

The skill does not request always: true or other elevated platform privileges, and contains no code that modifies other skills or global agent settings. It requires agent-browser to be running, which is normal for browser automation tools.

What to consider before installing

This repository largely implements what it claims: a reader that uses agent-browser to copy/paste Tencent Docs content. However, two practical red flags: (1) scripts/check_weekly_report.py is a separate utility that hardcodes a specific Tencent Docs URL (DOC_URL) and will call read_sheet.py on that URL — this is unexpected for a generic reader and means the package includes logic targeting a particular document; (2) check_weekly_report.py reads environment variables QQ_USER_ID and WECOM_WEBHOOK_URL (not declared in the skill metadata) and will POST notifications to the webhook if configured. Before installing or running: - Inspect or remove scripts/check_weekly_report.py if you do not want automated checks against the hardcoded DOC_URL. - Do not set WECOM_WEBHOOK_URL unless you intend the skill to send messages; otherwise the script will skip notifications but still run the check. - Verify agent-browser is installed from the official npm registry and understand it will control a browser instance and interact with the clipboard (it can access open pages/tabs). - Run first in a sandboxed account or isolated environment and test with a disposable Tencent Docs URL. - If you only need on-demand reads, prefer calling scripts/read_sheet.py directly with your URL and avoid enabling any scheduled/automation that runs check_weekly_report.py. If you want me to, I can produce a trimmed version of the skill (remove or sanitize check_weekly_report.py), or walk through the exact lines that perform webhook calls and the hardcoded URL.

Tencent Docs Reader（腾讯文档读取器）📄

Author: Cm-wenge | License: MIT

读取腾讯文档在线表格内容，通过 agent-browser 的复制粘贴技巧绕过 Canvas 渲染。

Read content from Tencent Docs (腾讯文档) spreadsheets using agent-browser's copy-paste trick.

为什么需要这个工具？/ Why This Skill?

当你把腾讯文档的共享链接发给大模型（如 ChatGPT、Claude、GLM），模型无法直接读取文档内容——它只能看到一个 URL，无法"打开"链接去查看里面的表格数据。

When you share a Tencent Docs link with an LLM (ChatGPT, Claude, GLM, etc.), the model cannot read the document content — it only sees a URL and has no way to "open" and view the spreadsheet inside.

这个工具解决的问题：

This tool solves the problem by:

• 让 AI Agent 能读取腾讯文档表格数据 — 自动打开链接、提取内容、返回结构化文本
• 绕过 Canvas 渲染 — 腾讯文档用 Canvas 画表格，传统网页抓取方式（DOM解析、accessibility tree）全部失效，本工具通过浏览器复制粘贴法成功提取
• 无需授权登录 — 不需要腾讯账号登录，只需文档设置为"所有人可读"即可读取，适合自动化场景
• 适用于自动化场景 — 定时读取周报表格、数据汇总、内容监控等

典型场景 / Typical use cases:

用户：帮我看看这个腾讯文档里谁还没交周报 https://docs.qq.com/sheet/xxx
Agent：（调用本工具读取表格）→ 解析内容 → 告诉你结果

原理 / How It Works

腾讯文档用 Canvas 渲染表格，accessibility tree 读不到内容。这个工具通过以下方式绕过：

Tencent Docs renders tables with Canvas (unreadable by accessibility tree). This skill bypasses it by:

1. 在 agent-browser 中打开文档 / Opens the spreadsheet in agent-browser
2. 如指定了 --tab，先切换到对应子表 / Switches to the specified tab
3. 点击表格区域获取焦点 / Clicks the table area to focus
4. Ctrl+A 全选 → Ctrl+C 复制 / Select all → Copy
5. 打开空白页，创建 textarea / Opens a blank page with a textarea
6. Ctrl+V 粘贴 → eval 读取内容 / Paste → Read value via eval
7. 关闭临时标签页，返回文本内容 / Closes temp tab, returns tab-separated text

用法 / Usage

python {baseDir}/scripts/read_sheet.py --url "https://docs.qq.com/sheet/XXXX" --tab "SheetName"

参数 / Options

参数	必填	说明
--url	✅	腾讯文档表格 URL / Tencent Docs spreadsheet URL
--tab	❌	子表名称（如 "0328"），不填则读取当前活动子表 / Tab name
--auto-tab	❌	自动选择本周的周报标签页（本周五），找不到则选择当前日期前后两天内的标签页
--output	❌	保存到文件，不指定则输出到 stdout / Save to file

--auto-tab 逻辑说明：

1. 优先选择本周五的标签页（如0410）
2. 如果找不到，选择当前日期前后两天内的标签页（如4月11日，范围0409~0413）
3. 如果都找不到，返回失败（退出码1），调用者可据此发送通知

示例 / Examples

# 读取指定子表 / Read a specific tab
python {baseDir}/scripts/read_sheet.py --url "https://docs.qq.com/sheet/YOUR_SHEET_ID" --tab "0328"

# 自动选择本周的周报标签页（本周五） / Auto-select current week tab (this Friday)
python {baseDir}/scripts/read_sheet.py --url "https://docs.qq.com/sheet/YOUR_SHEET_ID" --auto-tab

# 保存到文件 / Save to file
python {baseDir}/scripts/read_sheet.py --url "https://docs.qq.com/sheet/YOUR_SHEET_ID" --tab "0328" --output result.txt

# 读取当前活动子表 / Read active tab
python {baseDir}/scripts/read_sheet.py --url "https://docs.qq.com/sheet/YOUR_SHEET_ID"

依赖 / Requirements

• Python 3.x — 运行读取脚本 / Reader script runtime
• agent-browser — 全局安装（npm install -g agent-browser）并启动守护进程（agent-browser start）

限制 / Limitations

• 仅支持在线表格（spreadsheets），不支持在线文档（docs）和幻灯片（slides）
• 需要文档设置为"所有人可读" — 本技能无需登录，但要求文档共享权限为"所有人可读"，需要登录或权限受限的文档无法读取
• 超大表格可能较慢（一次性复制全部内容）
• 腾讯文档前端改版可能导致失效，需适时更新

输出格式 / Output Format

制表符分隔，每行一条记录 / Tab-separated values, one row per line:

姓名    年龄    城市
张三    30      北京
李四    25      上海

可用 Python csv.reader（delimiter=\t）或标准文本工具解析。

---

Install via ClawHub (coming soon)

npx clawhub install tencent-docs-reader