Web Intel

网络搜索与信息提取统一入口。整合 content-extraction、Jina Reader、Firecrawl、web-access CDP。自动选最经济工具链。可被 BMW（brand-marketing-workflow）、wealth 子代理等直接调用。

Security Scan

The skill's code, instructions, and resource usage are consistent with a routing/search/extraction utility; it integrates local tools (firecrawl, web-access CDP) and r.jina.ai as described and does not request unrelated secrets or unusual installs.

Purpose & Capability

Name/description (web search + extraction router) match the implementation: it routes to r.jina.ai (Jina Reader), firecrawl CLI, a local web-access CDP proxy, and delegates to a content-extraction handler. No unrelated credentials, binaries, or surprising capabilities are requested.

Instruction Scope

SKILL.md and the CLI script operate within expected boundaries: they call firecrawl, r.jina.ai, and a localhost CDP proxy and import a local content-extraction router. Note that deep mode will access the local CDP (localhost:3456) which can expose pages requiring the user's browser login state — this is expected for the described purpose but is a privacy consideration.

Install Mechanism

No install spec (instruction-only + Python script). No downloads or archive extraction. The skill expects external tools to already exist (firecrawl, web-access). This is low-risk from an install perspective.

Credentials

The skill does not request environment variables, keys, or config paths. It does access local filesystem paths inside the skills workspace and contacts r.jina.ai and localhost:3456 — all proportional to a web-extraction router.

Persistence & Privilege

always is false and the skill does not request permanent platform-wide privileges or modify other skills. It runs when invoked and can be called autonomously by the agent (normal default).

Assessment

This skill appears coherent with its stated role as a router between local tools and r.jina.ai. Before enabling it: (1) Confirm you trust the installed firecrawl binary and understand what network calls it performs, since the script invokes it via subprocess. (2) Be aware that deep mode will use a local CDP proxy (localhost:3456) and therefore can read pages that require your browser's login/session state — only enable deep-mode usage if you trust the web-access CDP proxy and the skill. (3) Verify the presence and integrity of the referenced skills/content-extraction scripts in your workspace; if those are missing the skill will degrade but still run other backends. (4) If you have strong privacy requirements, test the script in a sandboxed environment first to observe network calls and outputs.

Web Intel — 统一网络检索层

职责边界

本技能是路由层，不重写任何已有工具。调度链：

web-intel
  ├── content-extraction/scripts/extract_router.py  (URL 分类)
  ├── r.jina.ai                                       (轻量全文提取)
  ├── firecrawl CLI                                   (搜索 + 抓取)
  └── web-access CDP Proxy (localhost:3456)           (登录态/反爬)

三档模式

模式	场景	Token 消耗	延迟
fast	只需标题+摘要，快速定位	~200-500	<2s
standard	需完整页面正文	~500-2000	2-8s
deep	登录内容 / JS重渲染 / 反爬站点	~2000-8000	10-30s

决策树

A. 搜索任务（给关键词）

[fast]    firecrawl search "query" --limit 5
          → 返回 title + url + snippet

[standard] firecrawl search + Jina 提取 top-1 全文
           → tool: firecrawl_search+jina

[deep]    firecrawl search --scrape --limit 5
          → 反爬/登录页面可升级到 web-access CDP

B. 提取任务（给 URL）

Step 0: extract_router.py 判断 URL 类型
  → 微信/飞书/YouTube → 委托 skills/content-extraction（专用 handler）
  → 通用网页 → 继续

[fast]    curl https://r.jina.ai/<url>
[standard] Jina 优先；失败 → firecrawl scrape
[deep]    web-access CDP（localhost:3456）优先；降级 Jina → firecrawl scrape

C. 证券/财经（--type finance）

fast:     firecrawl search "$TICKER 财报/行情/股价" --limit 5
standard: 同 fast + Jina 提取 top-1（东方财富/雪球）
deep:     → skills/stock-research-engine（完整基本面分析）

D. 竞品研究（--type competitor，BMW 使用）

fast:     firecrawl search "品牌名 营销/产品/用户反馈" --limit 5
standard: fast + Jina 提取各结果全文
deep:     firecrawl search --scrape；需要 CDP 的站点走 web-access

调用约定（供其他技能/子代理）

# 搜索
python3 ~/.openclaw/workspace/skills/web-intel/scripts/web_intel.py \
  --query "比亚迪Q1财报" --mode fast --type finance

# URL 提取
python3 ... --url https://example.com --mode standard

# 竞品研究（BMW 调用）
python3 ... --query "小米SU7营销策略" --mode standard --type competitor

标准输出（JSON stdout）：

{
  "query": "...",
  "mode": "fast",
  "type": "finance",
  "tool_used": "firecrawl_search",
  "results": [{"title": "...", "url": "...", "snippet": "..."}],
  "full_content": null,
  "web_access_available": true,
  "latency_ms": 1200
}

web-access CDP 集成说明

web-access 通过 CDP Proxy（localhost:3456）直连用户 Chrome，天然携带登录态。

启动 CDP Proxy（deep 模式前置）：

bash ~/.openclaw/workspace/skills/web-access/scripts/check-deps.sh

web-intel 在 deep 模式下自动检测 CDP 可用性（web_access_available 字段）。CDP 可用时优先用于提取；不可用时降级到 Jina/Firecrawl。

不包含的能力（直接引用现有技能）

需求	使用技能
微信公众号提取	skills/content-extraction（handler=browser）
飞书文档提取	skills/content-extraction（handler=feishu）
YouTube 转录	skills/content-extraction（handler=transcript）
浏览器交互/表单	skills/browser + browser-use
个股深度研究	skills/stock-research-engine
本地记忆搜索	skills/search-memory
CDP 站点经验库	skills/web-access/references/site-patterns/