跳到正文
cccskills
免费在 GitHub 公开

web-search

Searches the open web for docs, comparisons, and news via You.com MCP or WebSearch. Use when a topic needs knowledge beyond GitHub, forums, or academia.

中文简介正在准备中。当前显示原文说明。

集成工具GitHub

实测评分 未实测

查看安装方法

包含的文件(1)

  • SKILL.md3.2 KB

SKILL.md(原文)

安装前可以查看交给 AI 代理的指令内容。

Web Search

When To Use

  • Finding vendor documentation, standards, or official guides
  • Comparisons and benchmarks across products or approaches
  • News, release notes, and recent status for a technology
  • Standalone searches, or as one channel of a /tome:research session: the planner dispatches tome:web-searcher from medium depth, and the channel runs a positive control (RFC 2119) before its topic queries so its silence can count in the coverage verdict

When NOT To Use

  • Code implementations (use Skill(tome:code-search))
  • Community opinions (use Skill(tome:discourse))
  • Academic literature (use Skill(tome:papers))

Search the open web for pages about a topic. This is the general-knowledge channel: the pages no domain-specific channel targets.

Sources (Priority Order)

  1. You.com MCP server (you-search tool) when configured
  2. WebSearch tool as fallback (no configuration needed)

The You.com MCP server is optional. Without it the channel runs entirely on the built-in WebSearch tool, so a session never fails because the server is absent.

Setup (Optional)

To prefer You.com results, add the MCP server to your Claude Code config once:

# Keyless free profile
claude mcp add you --transport http https://api.you.com/mcp?profile=free

# Authenticated (higher rate limits; YDC_API_KEY from you.com)
claude mcp add you --transport http https://api.you.com/mcp \
  --header "Authorization: Bearer $YDC_API_KEY"

A key is only needed for the authenticated endpoint. The free profile needs no key and no account.

Workflow

  1. Build search queries using tome.channels.web.expand_web_queries()
  2. Run each query through the You.com MCP you-search tool if the server is configured, otherwise through WebSearch
  3. Parse results via parse_you_mcp_result() (raises ValueError on a response shape that is not a you-search envelope) or parse_websearch_result() (raises ValueError on a result with no URL)
  4. Rank via rank_web_findings()
  5. Return Finding objects

Failure Handling

  • If the You.com MCP tool errors or is unreachable, retry the query through WebSearch and record the fallback in the query log so the report can show the channel ran degraded rather than ok
  • If both paths fail for every query, report the failure explicitly rather than returning fabricated or empty findings

Exit Criteria

  • At least one web search query built and executed for the requested topic, through You.com MCP or WebSearch
  • Results parsed into Finding objects with title, URL, and a relevance score, marked channel="web"
  • Findings returned from rank_web_findings() sorted by relevance score descending
  • If the You.com path failed and WebSearch was used, the fallback is reported rather than passed off as the primary source
  • If no results are found for any query, this is reported explicitly rather than returning an empty list silently

评价

还没有评价。欢迎分享你的使用感受。

评论

无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。

还没有评论。来写第一条吧。

提交后会立即公开。请勿填写邮箱、电话等个人信息。URL 最多 2 个。

0 / 1000 字

此表单受 reCAPTCHA 保护,适用 Google 的隐私权政策和服务条款。

同一仓库的技能

Shapes turns action-first: action leads, steps numbered, state restated. Use for ADHD-friendly output. Do not use to trim tokens; use response-compression.

中文简介正在准备中。当前显示原文说明。

athola/claude-night-market3432026年10月24日 更新

Inverts burden of proof for code additions. Use when reviewing PRs, planning refactors, or running unbloat to challenge every addition's necessity.

中文简介正在准备中。当前显示原文说明。

athola/claude-night-market3432026年10月24日 更新

Tracks per-agent token usage and flags waste in parallel dispatch. Use when evaluating parallel agent efficiency or after a multi-agent run.

中文简介正在准备中。当前显示原文说明。

athola/claude-night-market3432026年10月24日 更新

Coordinates Claude agent teams via filesystem protocol. Use when orchestrating parallel agents with task dependencies. Do not use for single-agent tasks.

中文简介正在准备中。当前显示原文说明。

athola/claude-night-market3432026年10月24日 更新

Evaluates API surface design, consistency, and exemplar alignment. Use when reviewing public API changes or before releasing a new API surface.

中文简介正在准备中。当前显示原文说明。

athola/claude-night-market3432026年10月24日 更新

Selects architecture paradigm via research before scaffolding. Use when architecture is undecided and the choice needs justification and documentation.

中文简介正在准备中。当前显示原文说明。

athola/claude-night-market3432026年10月24日 更新

查看 athola 的全部技能

报告此技能的问题