
This is a smart search router that wraps opencli commands to pick the right search source based on what you're looking for. Instead of hardcoding commands, it reads live help output to avoid documentation drift. The core routing is simple: if you name a site, it uses that site. Otherwise it picks one AI source (grok, doubao, or gemini) first, then adds one or two specialized sources only if needed. It enforces hard limits to prevent search spam, like one call per AI source per question and max two calls for other sites. Every query ends with a search summary showing which sites were hit, what terms were used, and how many times. Good for when you want Claude to search across Chinese and English sources without burning through rate limits or making redundant calls.
$ npx -y skills add joeseesun/opencli-skill --skill qiaomu-smart-search --agent claude-codeInstalls into .claude/skills of the current project.
根据话题和场景,将查询路由到最佳的 opencli 搜索源。此 skill 的核心目标不是记忆命令,而是先定位数据源,再让 Agent 通过 opencli 自己读取实时帮助,避免文档漂移。
每次使用前,必须先做下面两步:
opencli list -f yamlstrategy、browser、domain选定站点后,必须再做下面两步:
opencli <site> -h 查看该站点有哪些子命令opencli <site> <command> -h 查看参数、输出列、策略不要在 skill 文档里硬编码参数或假设命令签名;以 opencli ... -h 的实时输出为准。
只使用这一条规则,不再维护多套优先级:
grok、doubao、gemini 三选一。把“单个用户问题”理解为同一意图链路下的一次问题求解;同一轮追问、澄清、补充条件,若核心问题未变,仍算同一题。
先建立一份站点调用台账。每次真正执行搜索命令后,立刻更新:
sitequerycountstatus计数规则:
opencli list -f yaml、opencli <site> -h、opencli <site> <command> -h 属于预检与帮助,不计入搜索次数opencli <site> ... 搜索/查询执行,计为该站点 1 次调用频率上限:
触发限频后的处理:
Sponsors