Documents

Alibabacloud Iqs Search

Real-time web search and URL reading with structured results backed by Aliyun IQS.

What it does

Two local Node.js scripts (`search.mjs`, `readpage.mjs`) that call Aliyun IQS APIs to perform web search and page content extraction. Search supports three engines (Generic, LiteAdvanced, Deep) with controls for time range, content depth (summary or mainText), and result count (1–10). Page reading returns markdown, plain text, or HTML, with optional main-article extraction and a stealth mode for blocked pages. Requires Node.js ≥ 18 and an ALIYUN_IQS_API_KEY configured via environment variable or `~/.alibabacloud/iqs/env`. The skill must be executed through these scripts; built-in web tools should not be used as substitutes.

When to use it

Looking up current news or time-sensitive facts
Extracting structured content from a specific URL
Running deeper research that needs long mainText or multi-hop reasoning
Filtering search results to a recent time window (day/week/month/year)

The skill document

alibabacloud-iqs-search

Prerequisites

Node.js >= 18.0.0 (scripts use native fetch API, no external npm dependencies)

When to Use

User asks for current/recent information
User provides a URL to read
Need to verify facts or get real-time data
Research tasks requiring multiple sources

Decision Tree

Step 1: Determine Operation Type

If user provides a URL → Use readpage
If user asks a question needing web info → Use search

Step 2: For Search Operations

Follow the best practices to determine parameter values. Use default values when uncertain:

engineType
timeRange
contents

Step 3: For Page Reading

Follow the best practices to determine parameter values. Use default values when uncertain:

format
extractArticle
stealthMode

CRITICAL: Execution Method

You MUST execute the scripts via bash command (e.g., node scripts/search.mjs ... or node scripts/readpage.mjs ...). Do NOT use your built-in web_search, WebFetch, or any other internal tools as substitutes. If the script fails, retry or report the error — do NOT fall back to built-in tools.

Parameters & Best Practices

Search Parameters

Parameter	Type	Required	Default	Description
`--query`	string	Yes	-	Search query (1-500 chars)
`--engineType`	string	No	`LiteAdvanced`	Search engine type
`--timeRange`	string	No	`NoLimit`	Time range filter
`--contents`	string	No	-	Type of return content
`--numResults`	int	No	`10`	Number of search results (1-10)

Search Best Practices

1. Query Optimization (--query)

Keep queries concise (< 30 chars for best results)
Use specific keywords, avoid stop words
For news: include time context in query

2. Engine Selection (--engineType)

The four engines differ significantly in latency, recall depth, content length, and cost. Pick the cheapest engine that meets the task — do NOT default to Deep; it is ~10× slower and 50× more expensive than the standard engine.

Engine	Avg RT	Result count	snippet	mainText	Advanced filters	Multilingual	Cost ratio (vs Standard)	Use case
`Generic`	~950ms	~10	~150 chars	≤3000 chars	✗	Medium	1×	General search, news/realtime, scene queries like weather (supports `city`/`ip`)
`LiteAdvanced`	~500ms	1-50	~500 chars	≤3000 chars	✓	Good	Lite-tier 1×	Default recommendation: low-latency semantic search; snippets are already rich enough
`Deep`	~6s	1-50	≤500 chars	≤50000 chars	✓	Good (CN/EN)	Vertical 50×	Complex multi-step reasoning, research reports, offline/Agent tasks needing deep browsing

Decision rules:

Default → LiteAdvanced: low latency + semantic search + snippet covers most Agent needs without needing extra mainText fetches.
Choose Generic when: the query is short and clearly informational (news, weather, simple facts), or when minimizing cost matters; also the only engine honoring city / ip for scene results (weather etc.).
Choose Deep ONLY when: the question is multi-hop / complex reasoning (FRAMES/BrowseComp-style), OR you need very long mainText (≤50000 characters, ~16× longer than other engines) for downstream LLM reasoning.
- ⚠️ Avoid Deep for: real-time chat, simple lookups, high-QPS scenarios — latency (~6s) and cost (50×) are prohibitive.

3. Time Range Selection (--timeRange)

NoLimit: Default when uncertain - engine optimizes based on query relevance
OneDay: Today only
OneWeek: Last 7 days
OneMonth: Last 30 days
OneYear: Last 365 days

4. Content Return (--contents)

mainText: Return full main text content - Use when detailed information is needed, such as technical documentation, research reports, or in-depth articles
summary: Return concise summary only - Use when a quick overview is sufficient, or when the page content is too large and token reduction is needed

5. Result Count (--numResults)

Control number of results returned (default: 10, range: 1-10)

ReadPage Parameters

Parameter	Type	Required	Default	Description
`--url`	string	Yes	-	Target page URL
`--format`	string	No	`markdown`	Return format
`--timeout`	number	No	`60000`	Total timeout in milliseconds
`--pageTimeout`	number	No	`15000`	Page load timeout in milliseconds
`--stealth`	number	No	`0`	Enable stealth mode (0 or 1)
`--extractArticle`	boolean	No	`false`	Extract main article content only

ReadPage Best Practices

1. Format Selection (--format)

markdown: Best for articles, preserves structure (default)
text: Best for data extraction
html: When structure analysis needed

2. Article Extraction (--extractArticle)

Enable for: blogs, news articles
Disable for: product pages, directories

3. Handling Failures (--timeout, --stealth)

If timeout: Retry with increased --timeout value
If blocked: Enable --stealth 1
If still fails: Report to user

Command Line Usage

Search Examples

Basic Search

node scripts/search.mjs --query "量子计算原理" --engineType LiteAdvanced

Real-time Information Search

node scripts/search.mjs --query "最新金融政策" --engineType Generic --timeRange OneWeek

Search with Results Limit

node scripts/search.mjs --query "www.aliyun.com" --engineType LiteAdvanced --numResults 3

Search with Full Content

node scripts/search.mjs --query "AI 法案" --engineType LiteAdvanced --contents mainText

Search with Summary Only

node scripts/search.mjs --query "人工智能行业年度报告" --engineType LiteAdvanced --contents summary

Deep Research Search (complex multi-hop / long mainText)

# Returns up to 50000-char mainText per result; latency ~6s. Default timeout auto-bumps to 60s for Deep.
node scripts/search.mjs --query "对比 GPT-5 与 Claude Opus 4.7 的代码能力差异" --engineType Deep --numResults 5 --contents mainText

ReadPage Examples

Page Reading with Markdown Format

node scripts/readpage.mjs --url "https://example.com/article" --format markdown --extractArticle true

Page Reading with Plain Text Format

node scripts/readpage.mjs --url "https://example.com/article" --format text --timeout 60000

Page Reading with Stealth Mode

node scripts/readpage.mjs --url "https://example.com/article" --format markdown --stealth 1 --extractArticle true

Output Verification

After executing any search.mjs or readpage.mjs command:

Check the exit code: If non-zero, the command failed — do not claim success.
Verify output exists: If you saved results to a file, run ls -la and head -20 to confirm the file exists and contains valid data.
Never fabricate results: If the command failed or returned an error, report the failure honestly. Do not generate content from your own knowledge and present it as search results.

Error Handling

ALIYUN_IQS_API_KEY Configuration Error

If the script returns an error about missing API key:

STOP the current task immediately. Do NOT fall back to built-in tools (WebFetch, web_search, curl, etc.) as substitutes.
Report the error to the user and ask the user to configure the API key:
Retry the task with following instruction: Method 1: Environment Variable

export ALIYUN_IQS_API_KEY="your-api-key"

Method 2: Configuration File Create or edit ~/.alibabacloud/iqs/env:

ALIYUN_IQS_API_KEY=your-api-key

Related skills

deepseek-web-search

DeepSeek's latest official web search tool, now built right into the server — powered by DeepSeek's own powerful reasoning engine. Crush your knowledge cutoff: ask about breaking news, live events, or anything happening right now, and DeepSeek's sharp understanding turns raw web results into a crisp, synthesized answer backed by real source URLs. No link dumps, no noise — just DeepSeek's intelligence doing the heavy lifting. Fresh, fast, and authoritative. Not for structured result lists.

2 stars