Web Search — 网页 搜索

网页 搜索

Use this 技能 for 网页 discovery.

Typical cases:

the user wants recent facts, news, releases, or announcements you need to find the right docs page, b记录 post, repo, or product page you need multiple external sources before summarizing or comparing them the user asks vague things like “find the official docs”, “look up current pricing”, or “see what changed recently”

If the user already gave an exact URL and wants its contents, use 网页-fetch instead.

Command

运行:

node ./搜索.mjs --查询 "<查询>"

Optional flags:

--num结果s --type auto|fast|deep --livecrawl fallback|preferred --上下文MaxCharacters

Example:

node ./搜索.mjs \ --查询 "Anthropic 模型 上下文 Protocol latest announcements" \ --num结果s 5 \ --type fast

How to form good queries

Turn the user's 请求 into a focused 搜索 查询 before 运行ning the 工具.

Prefer queries that include:

the exact product, company, 库, 框架, or topic name the aspect you need, such as pricing, release notes, docs, 迁移 图形界面de, API reference, benchmark, or comparison time qualifiers when relevant, like 2026, latest, recent, or a version number source hints when useful, such as site:docs.example.com or site:github.com

When needed, do multiple narrower 搜索es instead of one broad 搜索.

Examples:

next.js 应用 路由r caching docs site:nextjs.org openAI 响应s API pricing 2026 cloudflare workers durable objects 迁移 图形界面de site:github.com vercel AI sdk 工具 calling examples 搜索 strategy 启动 with a tight 查询. Review the returned sources and snippets. If 结果s are weak, refine the 查询 rather than repeating the same one. If you identify a promising URL that needs close inspection, follow up with 网页-fetch. For comparisons or re搜索 summaries, prefer gathering a few solid sources over many noisy ones. Choosing options

Use the defaults unless the task clearly calls for something else.

--type auto: good default --type fast: use for quick fact-finding and strAIghtforward discovery --type deep: use for harder re搜索 tasks where recall matters more than speed --num结果s: lower it for narrow queries, increase it when surveying a space --上下文MaxCharacters: increase only when you truly need more returned 上下文 --livecrawl preferred: use when freshness matters and you want more live data How to handle 输出

The 命令行工具 returns raw 搜索 上下文 from Exa.

After 搜索ing:

summarize the findings instead of dumping the raw 输出 unless the user asks for it cite or mention the most relevant sources clearly call out uncertAInty, stale-looking 结果s, or conflicts between sources if the user asked a precise question and the 搜索 输出 is still ambiguous, fetch one or two authoritative pages and inspect them with 网页-fetch 失败 handling

If the 搜索 fAIls or returns weak 结果s:

tighten or reframe the 查询 reduce scope to a specific vendor, site, product, or timeframe split a compound question into separate 搜索es tell the user plAInly if the source 质量 is weak or current in格式化ion is hard to 验证 Notes No API key is required. The 端点 mirrors opencode's Exa-backed 搜索 flow. This 技能 is for discovery first; pAIr it with 网页-fetch for detAIled page retrieval.