Intelligent Search Router

Route queries to the optimal opencli search source based on topic and scenario. The core goal of this skill is not to memorize commands, but to first locate the data source, then let the Agent read real-time help by itself through

opencli

to avoid document drift.

Mandatory Pre-check

You must complete the following two steps before each use:

Run
```
opencli list -f yaml
```
Use live registry to confirm whether the candidate site exists, and check
```
strategy
```
,
```
browser
```
,
```
domain
```

After selecting the site, you must complete the following two more steps:

Run
```
opencli <site> -h
```
to view the subcommands of the site
If a subcommand is locked, run
```
opencli <site> <command> -h
```
to view parameters, output columns, and policies

Do not hardcode parameters or assume command signatures in the skill documentation; take the real-time output of

opencli ... -h

as the standard.

Main Routing Rules

Only use this one rule, no multiple sets of priorities are maintained:

When the user explicitly specifies a website, platform or data source, use the corresponding website directly.
When the user does not specify a website, prioritize selecting only one AI source: choose one from
```
grok
```
,
```
doubao
```
,
```
gemini
```
.
When the content returned by AI is insufficient, lacks original data, requires authoritative proof or vertical results, supplement 1-2 dedicated sources.

Single Query Budget and Rate Limiting

Understand a "single user question" as a one-time problem solving under the same intent link; follow-up questions, clarifications, and supplementary conditions in the same round are still counted as the same question if the core problem remains unchanged.

First create a site call ledger. Update it immediately after each actual execution of the search command:

```
site
```
```
query
```
```
count
```
```
status
```

Counting rules:

```
opencli list -f yaml
```
,
```
opencli <site> -h
```
,
```
opencli <site> <command> -h
```
are pre-check and help operations, not counted in the number of searches
One actual execution of
```
opencli <site> ...
```
search/query counts as 1 call for the site
Failure of the same site due to error, timeout, verification code, anti-crawling, abnormal login status also counts as 1 call, do not retry infinitely

Rate limit:

Hard limit for AI sites: within the same question, each AI site can be called at most 1 time
The default policy is still to select only 1 AI site, do not string multiple AI sites into the regular process
Only when the user explicitly requires comparison of multiple AI sites can you call other AI sites additionally; but each named AI site can still be called at most 1 time
Non-AI sites can be called at most 2 times by default
The second call to a non-AI site must have a clear reason, for example, the first result is too broad, and you need to add time, region, category, sorting or keyword restrictions
Do not make a third call to non-AI sites; if the information is still insufficient, stop expanding the search and clearly explain the gap

Processing after rate limit is triggered:

Record: "Skipped: <site> has reached the rate limit"
Prioritize switching to other similar sites
If there is no suitable alternative source, answer directly based on the collected information, and explain the coverage and gap

Post-Query Report

After each query, a short "search summary" must be added at the end of the answer, including at least the following three items:

What websites were used for search
What terms were searched on each website
How many times each website was searched

If there are sites skipped due to rate limit, they should also be clearly written.

It is recommended to use the following fixed format:

Search Summary
- Site: <site1> | Query Term: <term1> | Times: <n>
- Site: <site2> | Query Terms: <term2>; <term3> | Times: <n>
- Skipped: <site3>, Reason: Reached rate limit

AI Source Selection

```
grok
```
Suitable for real-time discussions, English Internet public opinion, Twitter/X context, hot topic tracking.
```
doubao
```
Suitable for Chinese context, ByteDance Douyin ecosystem, lifestyle content, Chinese hot topics and pan-Chinese Q&A.
```
gemini
```
Suitable for global web pages, English materials, general information retrieval, background overview.

If the user does not specify a website, first judge the language and context by default, and then select only one from these three.

Once an AI site has executed a real query once, do not call the AI site again after rewriting keywords in the same question. If the answer is insufficient, prioritize supplementing dedicated sources, do not repeatedly call the same AI site.

AI Query Term Suggestions

When using AI sources, do not only provide an overly short keyword. Prioritize constructing queries in the form of "topic + goal + limiting conditions".

Topic The object, event, product, person, company, technical term that the user actually wants to query.
Goal What result is wanted, such as summary, comparison, reason, trend, recommendation, original clue.
Limiting conditions Language, region, time range, platform range, audience, price band, job location, whether to cite original sources.

Prioritize using the following expressions:

```
<topic> + <question you want to answer>
```
```
<topic> + <time range/region/language>
```
```
<topic> + <platform or source range>
```
```
<topic> + <output requirement>
```

Avoid entering only:

Single noun
Hot issues without time range
Shopping, job hunting, travel issues without regional restrictions
Social media issues without platform restrictions

Timing for Dedicated Source Supplement

Supplement dedicated sources when any of the following situations occur:

What AI gives is a summary, but you need original posts, original videos, original products or original job results
AI coverage is insufficient, missing vertical site information
Higher authority or stronger domain relevance is required
The user explicitly requires "find from a certain platform"

A single query is usually controlled at 1 AI source + 1 to 2 dedicated sources to avoid result overload.

Handling Unavailable Sources

When the site is unavailable:

Do not abort the entire search due to the failure of a single source
Record: "Skipped: <site> is unavailable"
Fall back to other sites of the same type, or fall back to an AI source
Always take the actual results of
```
opencli list -f yaml
```
and
```
opencli <site> -h
```
as the standard

Do not assume that any site is "absolutely available". Even for public sites, take the live help and execution results in the current environment as the standard.

Reference Documents

Read the corresponding files as needed:

references/sources-ai.md
— AI default sources
references/sources-tech.md
— Technology / Academic
references/sources-social.md
— Social media
references/sources-media.md
— Media / Entertainment
references/sources-info.md
— News / Knowledge
references/sources-shopping.md
— Shopping
references/sources-travel.md
— Travel
references/sources-other.md
— Other vertical sources

Only read files related to the current query, no need to load all.

qiaomu-smart-search

NPX Install

Tags

SKILL.md Content (Chinese)