deep-research

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Deep Research

深度研究

Analyze any topic, domain, or paper and generate a beautiful HTML report using Actionbook browser automation and json-ui rendering.

分析任意主题、领域或论文，并通过Actionbook浏览器自动化与json-ui渲染生成美观的HTML报告。

Usage

使用方法

/deep-research:analyze <topic>
/deep-research:analyze <topic> --lang zh
/deep-research:analyze <topic> --output ./reports/my-report.json

Or simply tell Claude: "帮我深度研究 XXX 并生成报告" / "Research XXX and generate a report"

/deep-research:analyze <主题>
/deep-research:analyze <主题> --lang zh
/deep-research:analyze <主题> --output ./reports/my-report.json

或者直接告诉Claude："帮我深度研究 XXX 并生成报告" / "Research XXX and generate a report"

Parameters

参数说明

Parameter	Required	Default	Description
`topic`	Yes	-	The subject to research (any text)
`--lang`	No	`both`	Language: `en` , `zh` , or `both` (bilingual)
`--output`	No	`./output/<topic-slug>.json`	Output path for JSON report

参数	是否必填	默认值	说明
`topic`	是	-	要研究的主题（任意文本）
`--lang`	否	`both`	语言选项： `en` （英文）、 `zh` （中文）或 `both` （双语）
`--output`	否	`./output/<topic-slug>.json`	JSON报告的输出路径

Topic Detection

主题检测

Pattern	Type	Strategy
`arxiv:XXXX.XXXXX`	Paper	arXiv Advanced Search (Step 2b) + ar5iv deep read
`doi:10.XXX/...`	Paper	Resolve DOI, then arXiv Advanced Search for related work
Academic keywords (paper, research, model, algorithm)	Academic topic	arXiv Advanced Search (Step 2b) + Google for non-academic sources
URL	Specific page	Fetch and analyze the page
General text	Topic research	Google search + arXiv Advanced Search if relevant

匹配模式	类型	策略
`arxiv:XXXX.XXXXX`	学术论文	arXiv高级搜索（步骤2b）+ ar5iv深度读取
`doi:10.XXX/...`	学术论文	解析DOI，然后通过arXiv高级搜索查找相关研究
学术关键词（paper、research、model、algorithm）	学术主题	arXiv高级搜索（步骤2b）+ 谷歌搜索非学术来源
URL	特定页面	获取并分析页面内容
通用文本	主题研究	谷歌搜索 + 相关时使用arXiv高级搜索

Architecture

架构

┌──────────┐     ┌──────────────┐     ┌──────────────┐     ┌──────────┐
│  Claude   │────▶│  Actionbook  │────▶│  Web Pages   │────▶│ Extract  │
│  Code     │     │  Browser CLI │     │  (multiple)  │     │ Content  │
└──────────┘     └──────────────┘     └──────────────┘     └─────┬────┘
      │                                                           │
      │          ┌──────────────┐     ┌──────────────┐           │
      ├─────────▶│  Actionbook  │     │ arXiv Adv.   │           │
      │          │  search/get  │────▶│ Search Form  │──────────▶│
      │          │  (selectors) │     │ (40+ fields) │           │
      │          └──────────────┘     └──────────────┘           │
      │                                                           │
      │    Actionbook indexes arXiv form selectors,               │
      │    enabling field-specific, filtered academic              │
      │    searches that WebFetch/WebSearch CANNOT do.             │
      │                                                           │
┌──────────┐     ┌──────────────┐     ┌──────────────┐           │
│  Open in │◀────│   json-ui    │◀────│  Write JSON  │◀──────────┘
│  Browser │     │   render     │     │  Report      │  Synthesize
└──────────┘     └──────────────┘     └──────────────┘

┌──────────┐     ┌──────────────┐     ┌──────────────┐     ┌──────────┐
│  Claude   │────▶│  Actionbook  │────▶│  Web Pages   │────▶│ Extract  │
│  Code     │     │  Browser CLI │     │  (multiple)  │     │ Content  │
└──────────┘     └──────────────┘     └──────────────┘     └─────┬────┘
      │                                                           │
      │          ┌──────────────┐     ┌──────────────┐           │
      ├─────────▶│  Actionbook  │     │ arXiv Adv.   │           │
      │          │  search/get  │────▶│ Search Form  │──────────▶│
      │          │  (selectors) │     │ (40+ fields) │           │
      │          └──────────────┘     └──────────────┘           │
      │                                                           │
      │    Actionbook indexes arXiv form selectors,               │
      │    enabling field-specific, filtered academic              │
      │    searches that WebFetch/WebSearch CANNOT do.             │
      │                                                           │
┌──────────┐     ┌──────────────┐     ┌──────────────┐           │
│  Open in │◀────│   json-ui    │◀────│  Write JSON  │◀──────────┘
│  Browser │     │   render     │     │  Report      │  Synthesize
└──────────┘     └──────────────┘     └──────────────┘

Why Actionbook, Not WebFetch/WebSearch?

为何选择Actionbook而非WebFetch/WebSearch？

Capability	Actionbook	WebFetch/WebSearch
Operate complex web forms (dropdowns, checkboxes, date pickers)	Yes — uses indexed selectors	No
arXiv: search by Author, Title, Abstract separately	Yes — `#terms-0-field` select	No — keyword only
arXiv: filter by subject (CS, Physics, Math, ...)	Yes — category checkboxes	No
arXiv: filter by date range or specific year	Yes — date inputs	No
Read pages with verified selectors (no guessing)	Yes — `actionbook get`	No — raw HTML parse
Interact with any indexed site's UI	Yes — click, type, select	No — read-only

This is the core value of Actionbook for research: it turns web forms into structured, programmable interfaces for AI agents.

功能	Actionbook	WebFetch/WebSearch
操作复杂网页表单（下拉框、复选框、日期选择器）	是 — 使用索引化选择器	否
arXiv：按作者、标题、摘要单独搜索	是 — 通过 `#terms-0-field` 选择	否 — 仅支持关键词搜索
arXiv：按学科筛选（计算机科学、物理学、数学等）	是 — 分类复选框	否
arXiv：按日期范围或特定年份筛选	是 — 日期输入框	否
通过验证后的选择器读取页面内容（无需猜测）	是 — 使用 `actionbook get`	否 — 仅支持原始HTML解析
与任意索引化网站的UI交互	是 — 支持点击、输入、选择操作	否 — 仅支持只读访问

这正是Actionbook用于研究的核心价值：它将网页表单转化为AI Agent可调用的结构化、可编程接口。

MUST USE Actionbook CLI

必须使用Actionbook CLI

Always use
actionbook browser
commands for web browsing. Never use WebFetch or WebSearch.

bash

actionbook browser open <url>          # Navigate to page
actionbook browser snapshot            # Get accessibility tree
actionbook browser text [selector]     # Extract text content
actionbook browser screenshot [path]   # Capture visual
actionbook browser click <selector>    # Click element
actionbook browser close               # Close browser (ALWAYS do this at end)

始终使用
actionbook browser
命令进行网页浏览，切勿使用WebFetch或WebSearch。

bash

actionbook browser open <url>          # 导航至页面
actionbook browser snapshot            # 获取可访问性树
actionbook browser text [selector]     # 提取文本内容
actionbook browser screenshot [path]   # 捕获截图
actionbook browser click <selector>    # 点击元素
actionbook browser close               # 关闭浏览器（务必在最后执行）

Complete Workflow

完整工作流程

Step 1: Plan Search Strategy

步骤1：规划搜索策略

Based on the topic, generate 5-8 search queries from different angles:

Core definition / overview
Latest developments / news
Technical details / implementation
Comparisons / alternatives
Expert opinions / analysis
Use cases / applications

Search order — ALWAYS query Actionbook API first, then search:

Step	Action	Why
Step 2 (FIRST)	Query Actionbook API	Get verified selectors for arXiv Advanced Search form, ar5iv papers, and any other known sites BEFORE browsing. This is the foundation for all subsequent steps.
Step 3 (SECOND)	arXiv Advanced Search	Use Actionbook selectors from Step 2 to perform multi-field, filtered academic search. Even non-academic topics often have relevant papers.
Step 4 (THIRD)	Google / Bing search	Supplement with blogs, news, code, discussions, non-academic sources.

IMPORTANT: Always query Actionbook API first (Step 2) to get selectors, then use them in arXiv Advanced Search (Step 3). This is what makes Actionbook-powered research fundamentally different from WebFetch/WebSearch — the agent knows the exact selectors for every form field before it even opens the browser.

根据主题，从不同角度生成5-8个搜索查询：

核心定义/概述
最新进展/新闻
技术细节/实现方案
对比分析/替代方案
专家观点/深度分析
应用场景/实际案例

搜索顺序 — 始终先调用Actionbook API，再进行搜索：

步骤	操作	原因
步骤2（优先）	调用Actionbook API	在浏览前获取arXiv高级搜索表单、ar5iv论文及其他已知网站的验证选择器，这是后续所有步骤的基础。
步骤3（其次）	arXiv高级搜索	使用步骤2获取的Actionbook选择器执行多字段、多条件的学术搜索。即使是非学术主题，通常也有相关论文可供参考。
步骤4（最后）	谷歌/必应搜索	补充博客、新闻、代码、讨论等非学术来源的信息。

重要提示： 始终先调用Actionbook API（步骤2）获取选择器，再用于arXiv高级搜索（步骤3）。这正是基于Actionbook的研究与WebFetch/WebSearch的根本区别 — Agent在打开浏览器前就已知道每个表单字段的确切选择器。

Step 2: Query Actionbook API for Selectors (ALWAYS DO THIS FIRST)

步骤2：调用Actionbook API获取选择器（务必优先执行）

BEFORE browsing any URL, query Actionbook's indexed selectors. This gives you verified CSS/XPath selectors instead of guessing.

bash

undefined

在浏览任何URL之前，先查询Actionbook的索引化选择器。 这样可以获取经过验证的CSS/XPath选择器，无需猜测。

bash

undefined

Search for indexed actions by domain

按域名搜索索引化操作

actionbook search "<keywords>" -d "<domain>"

actionbook search "<关键词>" -d "<域名>"

Get detailed selectors for a specific page

获取特定页面的详细选择器

actionbook get "<domain>:/<path>:<area>"


**Pre-indexed sites useful for research:**

| Site | area_id | Key Selectors |
|------|---------|---------------|
| arXiv Advanced Search | `arxiv.org:/search/advanced:default` | **40+ selectors**: field select, term input, category checkboxes (CS/Physics/Math/...), date range filters, cross-list control — used in Step 3 |
| ar5iv paper | `ar5iv.labs.arxiv.org:/html/{paper_id}:default` | `h1.ltx_title_document` (title), `div.ltx_authors` (authors), `div.ltx_abstract` (abstract), `section.ltx_section` (sections) |
| Google Scholar | `scholar.google.com:/:default` | `#gs_hdr_tsi` (search input), `#gs_hdr_tsb` (search button) |
| arXiv homepage | `arxiv.org:/:default` | Global search across 2.4M+ articles |

**For any URL you plan to visit**, run `actionbook search "<keywords>" -d "<domain>"` to check if it's indexed. Use indexed selectors when available; fall back to `actionbook browser snapshot` for unindexed sites.

**Example: Get arXiv Advanced Search selectors before searching:**

```bash

actionbook get "<域名>:/<路径>:<区域>"


**对研究有用的预索引网站：**

| 网站 | 区域ID | 关键选择器 |
|------|---------|---------------|
| arXiv高级搜索 | `arxiv.org:/search/advanced:default` | **40+个选择器**：字段选择、术语输入、分类复选框（计算机科学/物理学/数学等）、日期范围筛选、跨列表控制 — 用于步骤3 |
| ar5iv论文 | `ar5iv.labs.arxiv.org:/html/{paper_id}:default` | `h1.ltx_title_document`（标题）、`div.ltx_authors`（作者）、`div.ltx_abstract`（摘要）、`section.ltx_section`（章节） |
| 谷歌学术 | `scholar.google.com:/:default` | `#gs_hdr_tsi`（搜索输入框）、`#gs_hdr_tsb`（搜索按钮） |
| arXiv首页 | `arxiv.org:/:default` | 可搜索240万+篇文章的全局搜索功能 |

**对于计划访问的任何URL**，先运行`actionbook search "<关键词>" -d "<域名>"`检查是否已被索引。如果有索引化选择器则优先使用；未索引的网站则回退到`actionbook browser snapshot`。

**示例：搜索前获取arXiv高级搜索选择器：**

```bash

Query Actionbook for arXiv form selectors

查询Actionbook获取arXiv表单选择器

actionbook get "arxiv.org:/search/advanced:default"

Returns 40+ selectors: #terms-0-field, #terms-0-term, #classification-computer_science, etc.

返回40+个选择器：#terms-0-field、#terms-0-term、#classification-computer_science等

undefined

undefined

Step 3: arXiv Advanced Search (Using Actionbook Selectors)

步骤3：arXiv高级搜索（使用Actionbook选择器）

Key differentiator: WebFetch/WebSearch can only do simple keyword searches. Actionbook has indexed the entire arXiv Advanced Search form with 40+ verified selectors (queried in Step 2), enabling multi-field, multi-criteria academic searches — just like a human researcher would use the form.

Using the selectors obtained from Step 2, the Agent can:

Capability	Actionbook Selector	WebFetch/WebSearch
Search by specific field (Title, Author, Abstract)	`#terms-0-field` select → choose field	Not possible
Add multiple search terms with boolean logic	`button "Add another term +"`	Not possible
Filter by subject (CS, Physics, Math, etc.)	`#classification-computer_science` checkbox	Not possible
Filter by date range	`#date-filter_by-3` radio + `#date-from_date` / `#date-to_date`	Not possible
Filter by specific year	`#date-filter_by-2` radio + `#date-year` input	Not possible
Include/exclude cross-listed papers	`#classification-include_cross_list-0/1` radio	Not possible
Control results display	`#size` select, `#abstracts-0/1` radio	Not possible

Example: Search for recent CS papers by a specific author:

bash

undefined

核心差异化优势： WebFetch/WebSearch仅支持简单的关键词搜索。Actionbook已索引整个arXiv高级搜索表单的40+个验证选择器（步骤2中获取），可执行多字段、多条件的学术搜索 — 就像人类研究员操作表单一样。

使用步骤2获取的选择器，Agent可以：

功能	Actionbook选择器	WebFetch/WebSearch
按特定字段搜索（标题、作者、摘要）	`#terms-0-field` 选择 → 选择字段	无法实现
添加带布尔逻辑的多个搜索术语	`button "Add another term +"`	无法实现
按学科筛选（计算机科学、物理学、数学等）	`#classification-computer_science` 复选框	无法实现
按日期范围筛选	`#date-filter_by-3` 单选框 + `#date-from_date` / `#date-to_date`	无法实现
按特定年份筛选	`#date-filter_by-2` 单选框 + `#date-year` 输入框	无法实现
包含/排除跨列表论文	`#classification-include_cross_list-0/1` 单选框	无法实现
控制结果显示	`#size` 选择框、 `#abstracts-0/1` 单选框	无法实现

示例：搜索特定作者的近期计算机科学论文：

bash

undefined

Open arXiv Advanced Search

打开arXiv高级搜索页面

actionbook browser open "https://arxiv.org/search/advanced"

1. Set search field to "Author" and type author name

1. 将搜索字段设置为“作者”并输入作者姓名

actionbook browser click "#terms-0-field" actionbook browser click "option[value='author']" actionbook browser type "#terms-0-term" "Yann LeCun"

2. Filter to Computer Science only

2. 仅筛选计算机科学领域

actionbook browser click "#classification-computer_science"

3. Restrict to past 12 months

3. 限制为过去12个月的论文

actionbook browser click "#date-filter_by-1"

4. Show abstracts in results

4. 在结果中显示摘要

actionbook browser click "#abstracts-0"

5. Submit search

5. 提交搜索

actionbook browser click "button:has-text('Search'):nth(2)"

6. Extract results

6. 提取结果

actionbook browser text "#main-container"


**Example: Search by title keywords in a date range:**

```bash
actionbook browser open "https://arxiv.org/search/advanced"

actionbook browser text "#main-container"


**示例：按标题关键词在指定日期范围内搜索：**

```bash
actionbook browser open "https://arxiv.org/search/advanced"

Search in "Title" field

在“标题”字段中搜索

actionbook browser click "#terms-0-field" actionbook browser click "option[value='title']" actionbook browser type "#terms-0-term" "large language model agent"

Date range: 2025-01 to 2026-02

日期范围：2025-01至2026-02

actionbook browser click "#date-filter_by-3" actionbook browser type "#date-from_date" "2025-01-01" actionbook browser type "#date-to_date" "2026-02-09"

Submit and extract

提交并提取结果

actionbook browser click "button:has-text('Search'):nth(2)" actionbook browser text "#main-container"

undefined

actionbook browser click "button:has-text('Search'):nth(2)" actionbook browser text "#main-container"

undefined

Step 4: Supplement with Google / Bing Search

步骤4：补充谷歌/必应搜索

After arXiv, use Google/Bing to find non-academic sources (blogs, news, docs, code, discussions):

bash

undefined

完成arXiv搜索后，使用谷歌/必应查找非学术来源（博客、新闻、文档、代码、讨论）：

bash

undefined

Search via Google

谷歌搜索

actionbook browser open "https://www.google.com/search?q=<encoded_query>" actionbook browser text "#search"

actionbook browser open "https://www.google.com/search?q=<编码后的查询词>" actionbook browser text "#search"

Or search via Bing

或必应搜索

actionbook browser open "https://www.bing.com/search?q=<encoded_query>" actionbook browser text "#b_results"


Parse the search results to extract URLs and snippets. Collect the top 5-10 most relevant URLs. For each discovered URL, query Actionbook API (Step 2 pattern) to check if the site is indexed before visiting.

actionbook browser open "https://www.bing.com/search?q=<编码后的查询词>" actionbook browser text "#b_results"


解析搜索结果以提取URL和摘要，收集最相关的5-10个URL。对于每个发现的URL，先按照步骤2的方式调用Actionbook API检查网站是否已被索引，再进行访问。

Step 5: Deep Read Sources

步骤5：深度读取来源内容

For each relevant URL, first query Actionbook API (same as Step 2) to check if the site is indexed, then use verified selectors:

bash

actionbook browser open "<url>"
actionbook browser text                # Full page text (fallback)
actionbook browser text "<selector>"   # Use Actionbook selector if indexed

For arXiv papers, try sources in this order (newer papers often fail on ar5iv):

bash

undefined

对于每个相关URL，先调用Actionbook API（与步骤2相同）检查网站是否已被索引，再使用验证后的选择器：

bash

actionbook browser open "<url>"
actionbook browser text                # 回退方案：提取整页文本
actionbook browser text "<selector>"   # 如果已索引，使用Actionbook选择器

对于arXiv论文，按以下顺序尝试来源（较新的论文常常无法在ar5iv正常显示）：

bash

undefined

1. Try ar5iv first (best structured selectors from Actionbook)

1. 优先尝试ar5iv（Actionbook提供最佳结构化选择器）

actionbook browser open "https://ar5iv.org/html/<arxiv_id>" actionbook browser text "h1.ltx_title_document" # Title actionbook browser text "div.ltx_authors" # Authors actionbook browser text "div.ltx_abstract" # Abstract

actionbook browser open "https://ar5iv.org/html/<arxiv_id>" actionbook browser text "h1.ltx_title_document" # 标题 actionbook browser text "div.ltx_authors" # 作者 actionbook browser text "div.ltx_abstract" # 摘要

NOTE: section.ltx_section often fails on newer papers — use "article" as fallback

注意：section.ltx_section在较新论文上常常失效 — 回退使用"article"

2. If ar5iv content is truncated (<5KB), fall back to arxiv abstract + other sources

2. 如果ar5iv内容被截断（<5KB），回退到arXiv摘要+其他来源

actionbook browser open "https://arxiv.org/abs/<arxiv_id>" actionbook browser text "main"

3. Supplement with HuggingFace model cards and GitHub READMEs for full details

3. 补充HuggingFace模型卡片和GitHub README以获取完整细节

actionbook browser open "https://huggingface.co/papers/<arxiv_id>" actionbook browser text "main"


**Key lesson:** Don't rely solely on ar5iv. Always cross-reference 3-4 sources for completeness.

**For Google Scholar** (indexed by Actionbook):

```bash
actionbook browser open "https://scholar.google.com"

actionbook browser open "https://huggingface.co/papers/<arxiv_id>" actionbook browser text "main"


**关键经验：** 不要仅依赖ar5iv。始终交叉参考3-4个来源以确保内容完整。

**对于谷歌学术**（已被Actionbook索引）：

```bash
actionbook browser open "https://scholar.google.com"

Type into search: use selector #gs_hdr_tsi

搜索输入：使用选择器#gs_hdr_tsi

actionbook browser click "#gs_hdr_tsi"

... type query, click #gs_hdr_tsb to search

... 输入查询词，点击#gs_hdr_tsb进行搜索


**For unindexed sites**, use snapshot to discover page structure:

```bash
actionbook browser open "<url>"
actionbook browser snapshot            # Get accessibility tree to find selectors
actionbook browser text "<discovered_selector>"


**对于未索引的网站**，使用snapshot发现页面结构：

```bash
actionbook browser open "<url>"
actionbook browser snapshot            # 获取可访问性树以查找选择器
actionbook browser text "<发现的选择器>"

Step 6: Synthesize Findings

步骤6：整合研究结果

Organize collected information into a coherent report:

Overview / Executive Summary
Key Findings
Detailed Analysis
Supporting Data / Evidence
Implications / Significance
Sources

将收集到的信息组织成连贯的报告：

概述/执行摘要
核心发现
详细分析
支持数据/证据
影响/重要性
信息来源

Step 7: Generate json-ui JSON Report

步骤7：生成json-ui JSON报告

Write a JSON file following the

@actionbookdev/json-ui

schema. Use the Write tool.

Output path:

./output/<topic-slug>.json

(or user-specified

--output

path)

按照

@actionbookdev/json-ui

架构编写JSON文件，使用Write工具。

输出路径：

./output/<topic-slug>.json

（或用户指定的

--output

路径）

Step 8: Render HTML

步骤8：渲染HTML

Try these in order until one works:

bash

undefined

按以下顺序尝试，直到成功：

bash

undefined

1. Local monorepo (if running inside actionbook project)

1. 本地单仓库（如果在actionbook项目中运行）

node packages/json-ui/dist/cli.js render <report.json> -o <report.html>

2. npx (if published)

2. npx（如果包已发布到npm）

npx @actionbookdev/json-ui render <report.json> -o <report.html>

3. npx with @latest

3. 使用@latest标签的npx

npx @actionbookdev/json-ui@latest render <report.json> -o <report.html>


If all fail, save the JSON file and inform the user of its path.

npx @actionbookdev/json-ui@latest render <report.json> -o <report.html>


如果所有方法都失败，保存JSON文件并告知用户文件路径。

Step 9: Open in Browser

步骤9：在浏览器中打开

bash

undefined

bash

undefined

macOS

open <report.html>

Linux

xdg-open <report.html>

undefined

xdg-open <report.html>

undefined

Step 10: Close Browser

步骤10：关闭浏览器

Always close the browser when done:

bash

actionbook browser close

完成后务必关闭浏览器：

bash

actionbook browser close

json-ui Report Template

json-ui报告模板

IMPORTANT: Always include BrandHeader and BrandFooter.

json

{
  "type": "Report",
  "props": { "theme": "auto" },
  "children": [
    {
      "type": "BrandHeader",
      "props": {
        "badge": { "en": "Deep Research Report", "zh": "深度研究报告" },
        "poweredBy": "Actionbook"
      }
    },
    {
      "type": "Section",
      "props": { "title": { "en": "Overview", "zh": "概述" }, "icon": "paper" },
      "children": [
        {
          "type": "Prose",
          "props": {
            "content": { "en": "English overview...", "zh": "中文概述..." }
          }
        }
      ]
    },
    {
      "type": "Section",
      "props": { "title": { "en": "Key Findings", "zh": "核心发现" }, "icon": "star" },
      "children": [
        {
          "type": "ContributionList",
          "props": {
            "items": [
              {
                "badge": { "en": "Finding", "zh": "发现" },
                "title": { "en": "...", "zh": "..." },
                "description": { "en": "...", "zh": "..." }
              }
            ]
          }
        }
      ]
    },
    {
      "type": "Section",
      "props": { "title": { "en": "Detailed Analysis", "zh": "详细分析" }, "icon": "bulb" },
      "children": [
        {
          "type": "Prose",
          "props": { "content": { "en": "...", "zh": "..." } }
        }
      ]
    },
    {
      "type": "Section",
      "props": { "title": { "en": "Key Metrics", "zh": "关键指标" }, "icon": "chart" },
      "children": [
        {
          "type": "MetricsGrid",
          "props": { "metrics": [], "cols": 3 }
        }
      ]
    },
    {
      "type": "Section",
      "props": { "title": { "en": "Sources", "zh": "信息来源" }, "icon": "link" },
      "children": [
        {
          "type": "LinkGroup",
          "props": { "links": [] }
        }
      ]
    },
    {
      "type": "BrandFooter",
      "props": {
        "timestamp": "YYYY-MM-DDTHH:MM:SSZ",
        "attribution": "Powered by Actionbook",
        "disclaimer": {
          "en": "This report was generated by AI using web sources. Verify critical information independently.",
          "zh": "本报告由 AI 基于网络来源生成，请独立验证关键信息。"
        }
      }
    }
  ]
}

重要提示：务必包含BrandHeader和BrandFooter。

json

{
  "type": "Report",
  "props": { "theme": "auto" },
  "children": [
    {
      "type": "BrandHeader",
      "props": {
        "badge": { "en": "Deep Research Report", "zh": "深度研究报告" },
        "poweredBy": "Actionbook"
      }
    },
    {
      "type": "Section",
      "props": { "title": { "en": "Overview", "zh": "概述" }, "icon": "paper" },
      "children": [
        {
          "type": "Prose",
          "props": {
            "content": { "en": "English overview...", "zh": "中文概述..." }
          }
        }
      ]
    },
    {
      "type": "Section",
      "props": { "title": { "en": "Key Findings", "zh": "核心发现" }, "icon": "star" },
      "children": [
        {
          "type": "ContributionList",
          "props": {
            "items": [
              {
                "badge": { "en": "Finding", "zh": "发现" },
                "title": { "en": "...", "zh": "..." },
                "description": { "en": "...", "zh": "..." }
              }
            ]
          }
        }
      ]
    },
    {
      "type": "Section",
      "props": { "title": { "en": "Detailed Analysis", "zh": "详细分析" }, "icon": "bulb" },
      "children": [
        {
          "type": "Prose",
          "props": { "content": { "en": "...", "zh": "..." } }
        }
      ]
    },
    {
      "type": "Section",
      "props": { "title": { "en": "Key Metrics", "zh": "关键指标" }, "icon": "chart" },
      "children": [
        {
          "type": "MetricsGrid",
          "props": { "metrics": [], "cols": 3 }
        }
      ]
    },
    {
      "type": "Section",
      "props": { "title": { "en": "Sources", "zh": "信息来源" }, "icon": "link" },
      "children": [
        {
          "type": "LinkGroup",
          "props": { "links": [] }
        }
      ]
    },
    {
      "type": "BrandFooter",
      "props": {
        "timestamp": "YYYY-MM-DDTHH:MM:SSZ",
        "attribution": "Powered by Actionbook",
        "disclaimer": {
          "en": "This report was generated by AI using web sources. Verify critical information independently.",
          "zh": "本报告由 AI 基于网络来源生成，请独立验证关键信息。"
        }
      }
    }
  ]
}

Paper Report Template (for arXiv papers)

论文报告模板（适用于arXiv论文）

When analyzing academic papers, use a richer template with:

```
PaperHeader
```
(title, arxivId, date, categories)
```
AuthorList
```
(authors with affiliations)
```
Abstract
```
(with keyword highlights)
```
ContributionList
```
(key contributions)
```
MethodOverview
```
(step-by-step method)
```
ResultsTable
```
(experimental results)
```
Formula
```
(key equations, LaTeX)
```
Figure
```
(paper figures from ar5iv)

分析学术论文时，使用更丰富的模板，包含：

```
PaperHeader
```
（标题、arxivId、日期、分类）
```
AuthorList
```
（带所属机构的作者列表）
```
Abstract
```
（带关键词高亮的摘要）
```
ContributionList
```
（核心贡献）
```
MethodOverview
```
（分步方法说明）
```
ResultsTable
```
（实验结果）
```
Formula
```
（关键公式，LaTeX格式）
```
Figure
```
（来自ar5iv的论文图表）

Available json-ui Components

可用的json-ui组件

Component	Use For	Key Props
`BrandHeader`	Report header	`badge` , `poweredBy`
`PaperHeader`	Paper metadata	`title` , `arxivId` , `date` , `categories`
`AuthorList`	Authors	`authors: [{name, affiliation}]` , `maxVisible`
`Section`	Major section	`title` , `icon` (paper/star/bulb/chart/code/link/info/warning)
`Prose`	Rich text	`content` (supports bold, italic, `code` , lists)
`Abstract`	Abstract text	`text` , `highlights: ["keyword"]`
`ContributionList`	Numbered findings	`items: [{badge, title, description}]`
`MethodOverview`	Step-by-step	`steps: [{step, title, description}]`
`MetricsGrid`	Key stats	`metrics: [{label, value, trend, suffix}]` , `cols`
`ResultsTable`	Data table	`columns` , `rows` , `highlights: [{row, col}]`
`Table`	Generic table	`columns: [{key, label}]` , `rows` , `striped` , `compact`
`Callout`	Info/tip/warning	`type` (info/tip/warning/important/note), `title` , `content`
`Highlight`	Blockquote	`type` (quote/important/warning/code), `text` , `source`
`KeyPoint`	Key finding card	`icon` , `title` , `description` , `variant`
`CodeBlock`	Code snippet	`code` , `language` , `title` , `showLineNumbers`
`Formula`	LaTeX equation	`latex` , `block` , `label`
`Figure`	Image(s)	`images: [{src, alt, width}]` , `label` , `caption`
`Image`	Single image	`src` , `alt` , `caption` , `width`
`DefinitionList`	Term/definition	`items: [{term, definition}]`
`LinkGroup`	Source links	`links: [{href, label, icon}]`
`Grid`	Grid layout	`cols` , children
`Card`	Card container	`padding` (sm/md/lg), `shadow`
`TagList`	Tags	`tags: [{label, color, href}]`
`BrandFooter`	Footer	`timestamp` , `attribution` , `disclaimer`

组件	用途	关键属性
`BrandHeader`	报告页眉	`badge` , `poweredBy`
`PaperHeader`	论文元数据	`title` , `arxivId` , `date` , `categories`
`AuthorList`	作者列表	`authors: [{name, affiliation}]` , `maxVisible`
`Section`	主要章节	`title` , `icon` （paper/star/bulb/chart/code/link/info/warning）
`Prose`	富文本内容	`content` （支持粗体、斜体、 `代码` 、列表）
`Abstract`	摘要文本	`text` , `highlights: ["keyword"]`
`ContributionList`	编号式发现	`items: [{badge, title, description}]`
`MethodOverview`	分步说明	`steps: [{step, title, description}]`
`MetricsGrid`	关键统计数据	`metrics: [{label, value, trend, suffix}]` , `cols`
`ResultsTable`	数据表格	`columns` , `rows` , `highlights: [{row, col}]`
`Table`	通用表格	`columns: [{key, label}]` , `rows` , `striped` , `compact`
`Callout`	信息/提示/警告	`type` （info/tip/warning/important/note）, `title` , `content`
`Highlight`	引用块	`type` （quote/important/warning/code）, `text` , `source`
`KeyPoint`	核心发现卡片	`icon` , `title` , `description` , `variant`
`CodeBlock`	代码片段	`code` , `language` , `title` , `showLineNumbers`
`Formula`	LaTeX公式	`latex` , `block` , `label`
`Figure`	图片组	`images: [{src, alt, width}]` , `label` , `caption`
`Image`	单张图片	`src` , `alt` , `caption` , `width`
`DefinitionList`	术语/定义列表	`items: [{term, definition}]`
`LinkGroup`	来源链接组	`links: [{href, label, icon}]`
`Grid`	网格布局	`cols` , children
`Card`	卡片容器	`padding` （sm/md/lg）, `shadow`
`TagList`	标签列表	`tags: [{label, color, href}]`
`BrandFooter`	报告页脚	`timestamp` , `attribution` , `disclaimer`

json-ui Known Pitfalls

json-ui已知陷阱

Pitfall	Symptom	Fix
`MetricsGrid.suffix` as i18n object	`text.replace is not a function`	`suffix` must be a plain string, not `{ "en": ..., "zh": ... }`
`MetricsGrid.value` as number	Render error	`value` must be a string (e.g., `"58.5"` not `58.5` )
Missing `BrandHeader` / `BrandFooter`	Report looks broken	Always include both
`Table` row values as i18n object	`[object Object]` in cells	Row cell values must be plain strings. Column `label` supports i18n, but row data does not. Use `"Runtimes / 运行时"` instead of `{ "en": "Runtimes", "zh": "运行时" }`
Very long Prose content	Truncated render	Split into multiple Prose blocks or use subsections

问题	症状	修复方案
`MetricsGrid.suffix` 使用国际化对象	`text.replace is not a function`	`suffix` 必须是纯字符串，不能是 `{ "en": ..., "zh": ... }`
`MetricsGrid.value` 使用数字类型	渲染错误	`value` 必须是字符串（例如 `"58.5"` 而非 `58.5` ）
缺少 `BrandHeader` / `BrandFooter`	报告显示异常	务必同时包含两者
`Table` 行值使用国际化对象	单元格显示 `[object Object]`	行单元格值必须是纯字符串。列 `label` 支持国际化，但行数据不支持。使用 `"运行时 / Runtimes"` 而非 `{ "en": "Runtimes", "zh": "运行时" }`
Prose内容过长	渲染被截断	拆分为多个Prose块或使用子章节

i18n Support

国际化支持

All text fields support bilingual output unless noted above:

json

{ "en": "English text", "zh": "中文文本" }

For

--lang en

, use plain strings. For

--lang zh

, use plain Chinese strings. For

--lang both

(default), use i18n objects.

Exceptions:

```
MetricsGrid
```
props
```
value
```
and
```
suffix
```
must always be plain strings.
```
Table
```
row cell values must be plain strings (column
```
label
```
supports i18n, but row data does not). For bilingual, use
```
"English / 中文"
```
format.

所有文本字段均支持双语输出除非上述特别说明：

json

{ "en": "English text", "zh": "中文文本" }

对于

--lang en

，使用纯字符串；对于

--lang zh

，使用纯中文字符串；对于默认的

--lang both

，使用国际化对象。

例外情况：

```
MetricsGrid
```
的
```
value
```
和
```
suffix
```
属性必须始终为纯字符串。
```
Table
```
的行单元格值必须为纯字符串（列
```
label
```
支持国际化，但行数据不支持）。如需双语，使用
```
"英文 / 中文"
```
格式。

Academic Paper Support

学术论文支持

arXiv Papers

arXiv论文

ar5iv.org HTML (preferred for reading, but often incomplete for papers < 3 months old):

Element	Selector (Actionbook-verified)	Reliability	Fallback
Title	`h1.ltx_title_document`	High	`div.ltx_abstract` includes title context
Authors	`div.ltx_authors`	High	—
Abstract	`div.ltx_abstract`	High	—
Full article	`article`	Medium	Use when section selectors fail
Sections	`section.ltx_section`	Low on new papers	`article` for all content
Section title	`h2.ltx_title_section`	Low on new papers	Parse from `article` text
Figures	`figure.ltx_figure`	Medium	—
Tables	`table.ltx_tabular`	Medium	—
Bibliography	`.ltx_bibliography`	Medium	—

Note: For papers submitted within the last ~3 months, ar5iv often renders incomplete content. Always check

actionbook browser text 2>&1 | wc -c

— if < 5KB, the page didn't fully render. Fall back to other sources.

arXiv API (for metadata via actionbook browser):

actionbook browser open "http://export.arxiv.org/api/query?id_list={arxiv_id}"
actionbook browser text

ar5iv.org HTML（优先用于阅读，但对于提交时间不足3个月的论文常常内容不完整）：

元素	Actionbook验证选择器	可靠性	回退方案
标题	`h1.ltx_title_document`	高	`div.ltx_abstract` 包含标题上下文
作者	`div.ltx_authors`	高	—
摘要	`div.ltx_abstract`	高	—
全文	`article`	中	当章节选择器失效时使用
章节	`section.ltx_section`	新论文可靠性低	使用 `article` 获取所有内容
章节标题	`h2.ltx_title_section`	新论文可靠性低	从 `article` 文本中解析
图表	`figure.ltx_figure`	中	—
表格	`table.ltx_tabular`	中	—
参考文献	`.ltx_bibliography`	中	—

注意： 对于提交时间约3个月内的论文，ar5iv常常无法完整渲染内容。始终检查

actionbook browser text 2>&1 | wc -c

— 如果结果小于5KB，说明页面未完全渲染，需回退到其他来源。

arXiv API（通过actionbook browser获取元数据）：

actionbook browser open "http://export.arxiv.org/api/query?id_list={arxiv_id}"
actionbook browser text

Recommended Source Priority for Papers

论文来源优先级推荐

Based on testing, use this priority order for maximum coverage:

Priority	Source	What you get	Reliability
1	`arxiv.org/abs/<id>`	Abstract, metadata, submission history	Very high
2	`huggingface.co/papers/<id>`	Abstract, community comments, related models/datasets	Very high
3	GitHub repo (from search results)	README with method details, model zoo, code	High
4	HuggingFace model card	Training recipe, benchmark results, quick start	High
5	`ar5iv.org/html/<id>`	Full paper HTML with structured selectors	Medium (fails on new papers)
6	Google Scholar / Semantic Scholar	Citations, related work	Medium

Key insight: Don't rely on a single source. The combination of arxiv abstract + HuggingFace + GitHub typically gives 90%+ of what you need, even when ar5iv fails.

基于测试结果，按以下优先级使用来源以获取最全面的信息：

优先级	来源	可获取内容	可靠性
1	`arxiv.org/abs/<id>`	摘要、元数据、提交历史	极高
2	`huggingface.co/papers/<id>`	摘要、社区评论、相关模型/数据集	极高
3	GitHub仓库（来自搜索结果）	包含方法细节、模型库、代码的README	高
4	HuggingFace模型卡片	训练方案、基准测试结果、快速开始指南	高
5	`ar5iv.org/html/<id>`	带结构化选择器的全文HTML	中（新论文常常失效）
6	谷歌学术/语义学术	引用文献、相关研究	中

核心见解： 不要依赖单一来源。结合arXiv摘要 + HuggingFace + GitHub通常能获取90%+的所需信息，即使ar5iv失效也不受影响。

Other Academic Sources

其他学术来源

Use

actionbook browser

to visit and extract content from:

Google Scholar (
```
scholar.google.com
```
) — Actionbook indexed, use
```
#gs_hdr_tsi
```
for search
Semantic Scholar (
```
semanticscholar.org
```
)
Papers With Code (
```
paperswithcode.com
```
)
Conference proceedings sites

使用

actionbook browser

访问并提取以下平台的内容：

谷歌学术（
```
scholar.google.com
```
）— 已被Actionbook索引，使用
```
#gs_hdr_tsi
```
进行搜索
语义学术（
```
semanticscholar.org
```
）
Papers With Code（
```
paperswithcode.com
```
）
会议论文集网站

Error Handling

错误处理

Error	Action
Browser fails to open	Run `actionbook browser status` , retry
Page load timeout (30s)	Skip source, try next. Common on papers.cool, slow academic sites
ar5iv content truncated (<5KB)	Paper too new for ar5iv. Fall back to arxiv abstract + HuggingFace + GitHub
`section.ltx_section` not found	ar5iv rendering incomplete. Use `actionbook browser text "article"` or `"main"` instead
Actionbook selector not found	Use `actionbook browser snapshot` to discover actual page structure
`actionbook search` returns no results	Site not indexed. Use `actionbook browser snapshot` to find selectors manually
json-ui render crash ( `text.replace` )	Check MetricsGrid `suffix` / `value` — must be plain strings, not i18n objects
`npx @actionbookdev/json-ui` 404	Package not on npm. Use local: `node packages/json-ui/dist/cli.js render`
No search results	Broaden search terms, try different angles
Render failed	Save JSON and inform user of the path

IMPORTANT: Always run

actionbook browser close

before finishing, even on errors.

错误	操作
浏览器无法打开	运行 `actionbook browser status` ，重试
页面加载超时（30秒）	跳过该来源，尝试下一个。在papers.cool等加载缓慢的学术网站上很常见
ar5iv内容被截断（<5KB）	论文太新，ar5iv尚未处理。回退到arXiv摘要 + HuggingFace + GitHub
`section.ltx_section` 未找到	ar5iv渲染不完整。使用 `actionbook browser text "article"` 或 `"main"` 替代
Actionbook选择器未找到	使用 `actionbook browser snapshot` 发现实际页面结构
`actionbook search` 无结果	网站未被索引。使用 `actionbook browser snapshot` 手动查找选择器
json-ui渲染崩溃（ `text.replace` 错误）	检查MetricsGrid的 `suffix` / `value` — 必须是纯字符串，不能是国际化对象
`npx @actionbookdev/json-ui` 返回404	包未发布到npm。使用本地方式： `node packages/json-ui/dist/cli.js render`
无搜索结果	放宽搜索词，尝试不同角度
渲染失败	保存JSON文件并告知用户文件路径

重要提示： 即使发生错误，完成后务必运行

actionbook browser close

。

Quality Guidelines

质量准则

Breadth: Research from at least 3-5 diverse sources
Depth: Read full articles, not just snippets
Accuracy: Cross-reference facts across sources
Structure: Use appropriate json-ui components for each content type
Attribution: Always include source links in the report
Freshness: Prefer recent sources when relevance is equal

广度：至少从3-5个不同来源进行研究
深度：阅读完整文章，而非仅摘要
准确性：跨来源交叉验证事实
结构：针对不同内容类型使用合适的json-ui组件
归因：报告中务必包含来源链接
时效性：当相关性相同时，优先选择较新的来源

Chinese Content Quality Guidelines (中文内容质量规范)

中文内容质量规范

CRITICAL: Chinese text must be written as native Chinese, NOT translated from English.

The

zh

field is not a translation — it is an independent Chinese version of the content. Write it as if authoring a Chinese tech article from scratch.

关键要求：中文文本必须以原生中文撰写，而非从英文直译。

zh

字段不是英文的翻译版本 — 它是内容的独立中文版本，需像从头撰写中文技术文章一样进行创作。

Common Problems to Avoid

需避免的常见问题

Problem	Bad Example	Good Example
被动语态过多	"Wasm 已被广泛采用"	"Wasm 已经大规模落地"
英语语序直译	"广泛但常常不可见地被采用"	"已经深入渗透到各类产品中，只是用户浑然不觉"
生硬术语拼接	"语言无关的异步通信"	"不绑定特定编程语言的异步通信机制"
直译英文短语	"关键缺失部分"	"最后一块拼图"
"地道"当形容词	"地道绑定"	"符合各语言习惯的绑定"
逗号长句（一逗到底）	"A，B，C，D，E。"	拆成 2-3 个短句
引用原文硬翻	"「许多用户并未意识到它正在被使用」"	用中文重新表述引用的核心意思，必要时保留原文人名

问题	错误示例	正确示例
被动语态过多	"Wasm 已被广泛采用"	"Wasm 已经大规模落地"
英语语序直译	"广泛但常常不可见地被采用"	"已经深入渗透到各类产品中，只是用户浑然不觉"
生硬术语拼接	"语言无关的异步通信"	"不绑定特定编程语言的异步通信机制"
直译英文短语	"关键缺失部分"	"最后一块拼图"
"地道"当形容词	"地道绑定"	"符合各语言习惯的绑定"
逗号长句（一逗到底）	"A，B，C，D，E。"	拆成2-3个短句
引用原文硬翻	"「许多用户并未意识到它正在被使用」"	用中文重新表述引用的核心意思，必要时保留原文人名

Writing Rules

写作规则

先写中文，再对照英文查漏：不要从英文逐句翻译。先理解内容本质，用中文重新组织表达。
主动语态优先：中文天然偏好主动句。"X 被 Y 采用" → "Y 采纳了 X" 或 "Y 已在用 X"。
短句 > 长句：中文读者偏好短句。超过 40 字的句子应拆分。
术语处理原则：
- 有公认中文译名的术语用中文：服务器 → 服务器、容器 → 容器、沙箱 → 沙箱
- 无公认译名的术语保留英文：Component Model、WASI、SIMD、Cold Start
- 首次出现时中英对照：「组件模型（Component Model）」，之后可只用中文
- 不要硬造中文译名（如"非干扰分析"，应写"Noninterference 分析"或解释性表述）
数据和事实保持一致：中英文的数字、日期、引用来源必须完全一致。
语气自然：可以用"说白了"、"换句话说"、"简单来说"等口语化连接词，避免全篇书面腔。
标点规范：用中文标点（，。；：「」）而非英文标点。引号优先用「」。

先写中文，再对照英文查漏：不要逐句翻译英文。先理解内容本质，用中文重新组织表达。
主动语态优先：中文天然偏好主动句。将"X 被 Y 采用"改写为"Y 采纳了 X"或"Y 已在用 X"。
短句优于长句：中文读者偏好短句。超过40字的句子应拆分。
术语处理原则：
- 有公认中文译名的术语使用中文：服务器、容器、沙箱
- 无公认译名的术语保留英文：Component Model、WASI、SIMD、Cold Start
- 首次出现时中英对照：「组件模型（Component Model）」，之后可仅用中文
- 不要硬造中文译名（如"非干扰分析"应写为"Noninterference 分析"或解释性表述）
数据和事实保持一致：中英文的数字、日期、引用来源必须完全一致。
语气自然：可以使用"说白了"、"换句话说"、"简单来说"等口语化连接词，避免全篇书面腔。
标点规范：使用中文标点（，。；：「」）而非英文标点。引号优先使用「」。

Bilingual Tone Reference

双语语气参考

English Tone	Chinese Equivalent	Example
"X has reached a critical milestone"	不要直译"关键里程碑"	"X 迎来了重要转折点" 或 "X 进入成熟期"
"This is the key missing piece"	不要直译"关键缺失部分"	"这是最后一块拼图" 或 "补上了最关键的短板"
"excels at cold starts"	不要直译"在冷启动方面表现卓越"	"冷启动速度远超同类方案"
"security scrutiny intensifies"	不要直译"安全审查加强"	"安全领域的关注度持续升温"
"emerging as a standard"	不要直译"正成为标准"	"逐渐确立了标准地位" 或 "大有一统江湖之势"
"widespread but invisible adoption"	不要直译"广泛但不可见的采用"	"已经悄然渗透到各类产品中"

英文语气	中文对应表达	示例
"X has reached a critical milestone"	不要直译"关键里程碑"	"X 迎来了重要转折点"或"X 进入成熟期"
"This is the key missing piece"	不要直译"关键缺失部分"	"这是最后一块拼图"或"补上了最关键的短板"
"excels at cold starts"	不要直译"在冷启动方面表现卓越"	"冷启动速度远超同类方案"
"security scrutiny intensifies"	不要直译"安全审查加强"	"安全领域的关注度持续升温"
"emerging as a standard"	不要直译"正成为标准"	"逐渐确立了标准地位"或"大有一统江湖之势"
"widespread but invisible adoption"	不要直译"广泛但不可见的采用"	"已经悄然渗透到各类产品中"