ByteTech

获取 ByteTech 技术文章的元信息、目录结构和正文内容。通过 Chrome DevTools MCP 直连用户本地 Chrome，复用登录态，抓包分析 API，提取文章数据。触发词：bytetech、字节技术、技术文章、bytetech article、获取 bytetech 文章、整理 bytetech。

Security Scan

The skill's purpose (fetch ByteTech articles by reusing a local Chrome session) is plausible, but the runtime instructions and bundled script assume/require several local CLIs and full access to the user's browser cookies without declaring them, and there are inconsistent tool/dependency declarations — this mismatch and the high-sensitivity of accessing local browser sessions warrant caution.

Purpose & Capability

The stated purpose — connect to the user's local Chrome via Chrome DevTools MCP to reuse login state and fetch article/meta/doc content — matches the commands and APIs used in SKILL.md and the script. However, the skill does not declare required binaries or credentials even though it clearly depends on tools (mcporter / chrome-devtools-mcp, agent-browser, node, optionally lark-cli). That omission is an incoherence: the skill will not work without those local tools, so the metadata understates its actual environmental needs.

Instruction Scope

Instructions explicitly instruct the agent to connect to the user's local Chrome and inherit cookies/session state, list and inspect XHR/fetch requests, evaluate scripts in page context, and fetch Lark (Feishu) document tokens and content. Those steps are within the stated goal (retrieving authenticated article content) but they inherently access sensitive artifacts (session cookies, internal APIs, user email/team info). The instructions also permit saving large responses to local files (/tmp), which could persist sensitive content. This is expected functionally but high-risk in sensitivity.

Install Mechanism

There is no install spec (instruction-only), which is low installation risk. But the shipped script and SKILL.md expect runtime tools (agent-browser, mcporter / chrome-devtools-mcp invoked via npx, node, lark-cli) and the declared allowed-tools only lists mcporter and an npx invocation, not agent-browser or node. The mismatch between used tools and declared allowed-tools / required binaries is an incoherence that could lead to unexpected failures or unauthorized use of other local tooling.

Credentials

The skill requests no environment variables or credentials in metadata, yet its operation depends on inheriting the user's browser cookies (gateway_sid, bytetech cookie, csrf token) and on accessing Feishu/Lark documents via tokens. It also references lark-cli usage which may require separate Feishu credentials. Asking to automatically inherit full browser session (cookies and auth) is functionally required but grants broad access to the user's authenticated web presence — a disproportionate sensitivity that users should explicitly consent to and understand.

Persistence & Privilege

always:false (no forced always-on) and normal model invocation settings are used. The skill does not request to modify other skills or system-wide config. The combination of autonomous invocation (platform default) plus the ability to access the local browser is potentially high-impact, but the skill does not request permanent presence.

What to consider before installing

This skill will connect to your local Chrome and reuse your logged-in session to fetch article metadata and Feishu (Lark) document contents. Before installing or running it: 1) Inspect the included script (scripts/bytetech-fetch.sh) and SKILL.md yourself — they will open pages and execute scripts in your browser context and can read cookies and auth-protected API responses. 2) Note the metadata does not declare required tools (agent-browser, mcporter / chrome-devtools-mcp, node, and optional lark-cli) and allowed-tools in SKILL.md omit agent-browser; ensure you understand and trust these CLIs. 3) If you are uncomfortable granting access to your browser session (which could expose other authenticated services), do not enable the skill or run it only in an isolated environment/profile. 4) Prefer running the script manually in a controlled terminal (not as an autonomous agent), and avoid persisting fetched outputs unless you intend to store sensitive internal content. 5) If you want to proceed, require explicit user consent each time the skill connects to the browser and avoid enabling autonomous invocation for untrusted skills.

ByteTech 文章获取

通过 Chrome DevTools MCP 连接用户本地 Chrome，复用登录态，抓取 ByteTech 技术文章。

前置条件

1. Chrome 已开启远程调试：chrome://inspect/#remote-debugging
2. mcporter 已配置 chrome-devtools server

验证连接

mcporter call chrome-devtools.list_pages

应返回 Chrome 中打开的页面列表。

获取文章详情（核心 API）

# 1. 导航到 bytetech 文章页（在 Chrome 新 tab 中打开）
mcporter call chrome-devtools.new_page url="https://bytetech.info/articles/<ARTICLE_ID>" timeout=30000

# 2. 刷新并抓包
mcporter call chrome-devtools.navigate_page type=reload

# 3. 过滤 XHR/Fetch 请求，找到文章详情 API
mcporter call chrome-devtools.list_network_requests resourceTypes='["xhr","fetch"]'
# 找 reqid 对应 /proxy_tech_api/v1/article/detail

# 4. 获取文章详情响应
mcporter call chrome-devtools.get_network_request reqid=<ID>

文章详情 API 返回完整元信息：

• data.article_info.title — 标题
• data.article_info.en_title — 英文标题
• data.article_info.summary — 摘要
• data.article_info.lark_doc_token — 飞书文档 token
• data.article_info.lark_doc_url — 飞书文档 URL
• data.article_info.view_cnt — 浏览量
• data.article_info.dig_cnt — 点赞数
• data.article_info.collect_cnt — 收藏数
• data.article_info.coin_cnt — 投币数
• data.article_info.labels — 标签列表 [{name, en_name, ...}]
• data.auther — 作者信息 [{name, full_name, email, department_name, ...}]

Request Body:

{"article_id":"<ARTICLE_ID>","id_type":8}

直接调用 bytetech API（无需浏览器刷新）

用 evaluate_script 在 bytetech 页面上下文中直接 fetch：

# 获取文章详情
mcporter call chrome-devtools.evaluate_script function='
async () => {
  const resp = await fetch("/proxy_tech_api/v1/article/detail", {
    method: "POST",
    headers: {"Content-Type": "application/json"},
    body: JSON.stringify({article_id: "7619471842051620907", id_type: 8})
  });
  return await resp.json();
}'

# 获取推荐文章
mcporter call chrome-devtools.evaluate_script function='
async () => {
  const resp = await fetch("/proxy_tech_api/v1/article/recommend", {
    method: "POST",
    headers: {"Content-Type": "application/json"},
    body: JSON.stringify({article_id: "7619471842051620907", id_type: 8, limit: 10})
  });
  return await resp.json();
}'

# 获取团队目录和关联文章
mcporter call chrome-devtools.evaluate_script function='
async () => {
  const resp = await fetch("/proxy_tech_api/v2/content/team_account/item_include_info", {
    method: "POST",
    headers: {"Content-Type": "application/json"},
    body: JSON.stringify({item_id: "7619471842051620907", id_type: 8})
  });
  return await resp.json();
}'

# 获取团队热门文章
mcporter call chrome-devtools.evaluate_script function='
async () => {
  const resp = await fetch("/proxy_tech_api/v1/content/team_account/rank/item", {
    method: "POST",
    headers: {"Content-Type": "application/json"},
    body: JSON.stringify({team_id: "<TEAM_ID>", limit: 10})
  });
  return await resp.json();
}'

获取文章正文

bytetech 文章正文存储在飞书文档中。

方式 A：lark-cli（推荐）

# 从 bytetech API 获取 lark_doc_token
# 然后用 lark-cli 获取飞书文档内容
lark-cli docs +fetch --as user --doc "<LARK_DOC_TOKEN>"

方式 B：Chrome DevTools MCP 导航到飞书文档

# 选择 bytetech 页面所在的 browser context（共享 cookies）
# 导航到飞书文档
mcporter call chrome-devtools.navigate_page type=url url="https://bytetech.info/articles/<ID>"
# 等待加载
mcporter call chrome-devtools.wait_for text='["文章"]' timeout=15000
# 提取页面文本
mcporter call chrome-devtools.take_snapshot

方式 C：evaluate_script 提取（从 SSR meta）

mcporter call chrome-devtools.evaluate_script function='
() => {
  const title = document.title.replace(/ - 文章 - ByteTech$/, "");
  const og = {};
  document.querySelectorAll("meta[property^=\"og:\"],meta[name]").forEach(m => {
    const k = m.getAttribute("property") || m.getAttribute("name");
    og[k] = m.content || "";
  });
  return JSON.stringify({title, description: og["og:description"] || og["description"]});
}'

搜索文章

# 在 bytetech 页面中搜索
mcporter call chrome-devtools.navigate_page type=url url="https://bytetech.info"
mcporter call chrome-devtools.wait_for text='["ByteTech"]' timeout=15000
mcporter call chrome-devtools.take_snapshot
# 找到搜索框 uid 后 fill
mcporter call chrome-devtools.fill uid="<SEARCH_UID>" value="AI Coding"
mcporter call chrome-devtools.press_key key=Enter
mcporter call chrome-devtools.wait_for text='["搜索结果"]' timeout=10000
mcporter call chrome-devtools.take_snapshot

抓包模式（接口探索）

# 导航到目标页面
mcporter call chrome-devtools.navigate_page type=reload

# 列出所有 API 请求
mcporter call chrome-devtools.list_network_requests resourceTypes='["xhr","fetch"]'

# 查看具体请求详情（含完整 headers 和 body）
mcporter call chrome-devtools.get_network_request reqid=<ID>

# 保存响应到文件（大响应）
mcporter call chrome-devtools.get_network_request reqid=<ID> responseFilePath=/tmp/response.json

API 速查表

API	Method	用途	Request Body
/proxy_tech_api/v1/article/detail	POST	文章详情（标题、摘要、标签、作者、飞书token）	{article_id, id_type:8}
/proxy_tech_api/v1/article/recommend	POST	推荐文章列表	{article_id, id_type:8, limit}
/proxy_tech_api/v2/content/team_account/item_include_info	POST	团队目录和关联文章	{item_id, id_type:8}
/proxy_tech_api/v1/content/team_account/rank/item	POST	团队热门文章排行	{team_id, limit}
/proxy_tech_api/v1/content/item_detail_inclusion	POST	文章详细关联信息	{item_id, id_type:8}
/proxy_tech_api/v2/label/tree	POST	标签树	{}
/proxy_tech_api/v2/search/hotwords	GET	热搜词	-
/proxy_tech_api/v1/login	POST	登录检查	{}
/api/v1/lark/component_auth	POST	飞书组件鉴权	{url}

Cookie 说明

bytetech 使用以下 Cookie 进行鉴权：

• gateway_sid — 网关 session
• bytetech=bytetech-user — 用户标识
• csrf_session_id — CSRF 防护

通过 Chrome DevTools MCP 的 --autoConnect 直连用户 Chrome，自动继承所有 Cookie，无需额外处理。

批量获取文章并总结

Step 1: 获取文章列表

先通过推荐 API 或团队排行 API 获取一批文章 ID：

# 推荐文章（基于某篇文章）
mcporter call chrome-devtools.evaluate_script function='
async () => {
  const resp = await fetch("/proxy_tech_api/v1/article/recommend", {
    method: "POST",
    headers: {"Content-Type": "application/json"},
    body: JSON.stringify({limit: 10})
  });
  const data = await resp.json();
  return JSON.stringify(data.data.map(i => ({
    article_id: i.article_info.article_id,
    title: i.article_info.title,
    author: i.auther.name,
    summary: (i.article_info.generated_summary || i.article_info.summary || "").substring(0, 200),
    lark_doc_token: i.article_info.lark_doc_token,
    views: i.article_info.view_cnt,
    likes: i.article_info.dig_cnt,
    collects: i.article_info.collect_cnt
  })));
}'

# 团队热门文章
mcporter call chrome-devtools.evaluate_script function='
async () => {
  const resp = await fetch("/proxy_tech_api/v1/content/team_account/rank/item", {
    method: "POST",
    headers: {"Content-Type": "application/json"},
    body: JSON.stringify({team_id: "<TEAM_ID>", limit: 10})
  });
  return await resp.json();
}'

Step 2: 批量获取文章详情

拿到 article_id 列表后，逐个调用 article/detail 获取完整元信息（标题、摘要、标签、作者、数据）。

Step 3: 按需获取正文

如果需要正文内容（不仅仅是摘要），逐个用 lark-cli 获取飞书文档：

lark-cli docs +fetch --as user --doc "<LARK_DOC_TOKEN>" 2>&1

注意：

• 每篇正文可能 10000-50000+ 字符，10 篇全文可能超过上下文窗口
• 建议策略：先只取摘要和 generated_summary 做初步筛选，再对感兴趣的 3-5 篇取全文
• 正文返回格式为 JSON，data.markdown 字段包含 markdown 格式的完整正文
• 长文的 markdown 中包含飞书自定义标签（如 <lark-td> 等），需要清洗

Step 4: 清洗正文（Markdown 清洗）

lark-cli 返回的飞书文档 markdown 包含大量飞书自定义标签，需要清洗：

lark-cli docs +fetch --as user --doc "<TOKEN>" 2>&1 | node -e "
let d='';process.stdin.on('data',c=>d+=c);process.stdin.on('end',()=>{
  const r=JSON.parse(d);
  let md = r.data.markdown;
  // 清洗飞书自定义标签
  md = md.replace(/<\/?lark-[a-zA-Z0-9_-]+>/g, '');
  md = md.replace(/<\/?grid[^>]*>/g, '');
  md = md.replace(/<\/?column[^>]*>/g, '');
  md = md.replace(/<mention-doc[^>]*>.*?<\/mention-doc>/g, '');
  md = md.replace(/<quote-container>.*?<\/quote-container>/gs, '');
  md = md.replace(/<image[^>]*\/>/g, '');
  // 保留标准 markdown 格式
  console.log(md);
})"

Step 5: 生成总结

根据获取到的数据，生成结构化总结，包括：

• 文章标题和链接：https://bytetech.info/articles/<article_id>
• 作者和部门
• 核心要点（从 summary 或正文中提取）
• 数据指标（阅读量、点赞、收藏）
• 趋势观察（多篇之间共同的话题方向）

链接格式

• ByteTech 文章：https://bytetech.info/articles/<article_id>
• 飞书文档：https://bytetech.info/articles/<article_id>#<lark_doc_token>

注意：ByteTech 文章需要飞书 OAuth 登录后才能查看，未登录会跳转登录页。

id_type 说明

• 8 — 文章
• 24 — 其他类型（视频等）