Install any skill in seconds. Free to start, no credit card required.
Get Started Free →【强制】所有技术文档查询必须使用本技能,禁止在主对话中直接使用 mcp__context7-mcp 工具。触发关键词:查询/学习/了解某个库或框架的文档、API用法、配置参数、错误解释、版本差异、代码示例、最佳实践。本技能通过 context7-researcher agent 执行查询,避免大量文档内容污染主对话上下文,保持 token 效率。
.claude/skills/aiskillstore-techdoc-search-and-orchestrate/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-12 | ✗→✓ | ▲ Improved | -36% | 0% |
| case-19 | ✗→✓ | ▲ Improved | -50% | 0% |
| case-20 | ✓→✓ | = Same ✓ | 66% | 0% |
| case-21 | ✓→✓ | = Same ✓ | 147% | 0% |
| case-22 | ✓→✓ | = Same ✓ | 21% | 0% |
本技能负责将技术文档查询任务委派给专业的 context7-researcher agent 执行,通过 agent 隔离来保持主对话上下文的清晰,避免 token 消耗污染。
识别需要技术文档查询的场景,并将任务委派给 context7-researcher agent,该 agent 专门使用 Context7 MCP 工具检索最新的技术文档。
当需要查询技术文档时,本技能负责将任务委派给 context7-researcher agent 执行,避免技术文档检索过程污染主对话上下文。
使用 Task tool 调用 context7-researcher agent:
Task tool 参数:
- subagent_type: "context7-researcher"
- description: 简短描述任务(3-5个字)
- prompt: 详细的查询需求当有多个技术文档查询需求时,可以选择以下两种方式:
方式一:顺序执行(单 Agent 多任务)
context7-researcher agent方式二:并行执行(多 Agent 并行)
context7-researcher agents选择建议:
示例对比:
用户问:"帮我查询 Supabase 的 realtime 如何对接和 Supabase 的 JS SDK 如何使用"
顺序执行方式:
调用 1 个 agent:
- subagent_type: "context7-researcher"
- description: "查询 Supabase 文档"
- prompt: "请完成以下查询:
1. 查询 Supabase realtime 的对接方法和示例
2. 查询 Supabase JS SDK 的使用方法和示例"并行执行方式(推荐):
同时调用 2 个 agents:
Agent 1:
- subagent_type: "context7-researcher"
- description: "查询 Supabase realtime"
- prompt: "查询 Supabase realtime 的对接方法和示例"
Agent 2:
- subagent_type: "context7-researcher"
- description: "查询 Supabase SDK"
- prompt: "查询 Supabase JS SDK 的使用方法和示例"这两个查询虽然都关于 Supabase,但是独立的主题(一个是 realtime 功能,一个是 SDK 使用),没有强依赖关系,应该优先选择并行执行以提高效率。
用户需求: "Supabase 怎么实现用户注册?"
执行方式:
调用 Task tool:
- subagent_type: "context7-researcher"
- description: "查询 Supabase 用户注册"
- prompt: "查询 Supabase 的用户注册 API 用法,包括代码示例"用户需求: "Next.js 的 App Router 怎么配置和 Server Actions 如何使用?"
执行方式:
并行调用 2 个 agents(推荐):
Agent 1:
- subagent_type: "context7-researcher"
- description: "查询 App Router 配置"
- prompt: "查询 Next.js App Router 的配置方法和示例"
Agent 2:
- subagent_type: "context7-researcher"
- description: "查询 Server Actions"
- prompt: "查询 Next.js Server Actions 的使用方法和示例"说明: 虽然都是 Next.js 的功能,但 App Router 配置和 Server Actions 是两个独立主题,没有强依赖,应优先并行查询。
用户需求: "我的 Next.js 项目报错 'Error: ENOENT: no such file or directory',这是什么原因?怎么解决?"
执行方式:
调用 1 个 agent 顺序执行:
- subagent_type: "context7-researcher"
- description: "排查 Next.js 错误"
- prompt: "请按顺序完成:
1. 先查询 Next.js 中 'ENOENT: no such file or directory' 错误的常见原因
2. 根据查到的原因,再查询对应的解决方案和最佳实践"说明: 这是典型的强依赖场景,必须先了解错误原因,才能针对性地查询解决方案,因此使用顺序执行。
用户需求: "Next.js 13 和 14 的路由有什么区别?"
执行方式:
调用 Task tool:
- subagent_type: "context7-researcher"
- description: "对比 Next.js 路由差异"
- prompt: "对比 Next.js 13 和 14 版本的路由系统差异,说明主要变更"说明: 这是单一查询任务,无需拆分,直接委派给一个 agent 执行。
通过本技能,主 agent 可以高效地将技术文档查询委派给专业 agent,保持对话流程清晰,优化 token 使用。
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | fail→fail | 14,524 | 2,121 | -85% | 1 | 1 | 0% | 2,731 | 1,760 | -36% | 0 | 0 | — |
case-13 | fail→fail | 20,819 | 6,089 | -71% | 1 | 1 | 0% | 3,670 | 2,212 | -40% | 0 | 0 | — |
case-14 | fail→fail | 18,529 | 4,327 | -77% | 1 | 1 | 0% | 3,471 | 1,880 | -46% | 0 | 0 | — |
case-15 | fail→fail | 21,094 | 11,484 | -46% | 1 | 1 | 0% | 3,513 | 2,025 | -42% | 0 | 0 | — |
case-06 | fail→fail | 14,928 | 6,577 | -56% | 1 | 1 | 0% | 2,480 | 1,777 | -28% | 0 | 0 | — |
case-03 | fail→fail | 18,503 | 8,022 | -57% | 1 | 1 | 0% | 3,461 | 1,803 | -48% | 0 | 0 | — |
case-04 | fail→fail | 17,313 | 3,470 | -80% | 1 | 1 | 0% | 3,063 | 1,601 | -48% | 0 | 0 | — |
case-01 | fail→fail | 18,395 | 2,342 | -87% | 1 | 1 | 0% | 3,670 | 1,727 | -53% | 0 | 0 | — |
case-02 | fail→fail | 18,694 | 10,217 | -45% | 1 | 1 | 0% | 3,113 | 1,960 | -37% | 0 | 0 | — |
case-07 | fail→fail | 20,836 | 5,960 | -71% | 1 | 1 | 0% | 3,665 | 2,057 | -44% | 0 | 0 | — |
case-08 | fail→fail | 12,034 | 6,977 | -42% | 1 | 1 | 0% | 2,212 | 1,767 | -20% | 0 | 0 | — |
case-09 | fail→fail | 12,709 | 7,036 | -45% | 1 | 1 | 0% | 2,557 | 1,701 | -33% | 0 | 0 | — |
case-10 | fail→fail | 18,138 | 3,552 | -80% | 1 | 1 | 0% | 3,379 | 1,723 | -49% | 0 | 0 | — |
case-11 | fail→fail | 15,191 | 2,187 | -86% | 1 | 1 | 0% | 2,986 | 1,679 | -44% | 0 | 0 | — |
case-12 | fail→pass | 15,536 | 10,665 | -31% | 1 | 1 | 0% | 2,977 | 1,903 | -36% | 0 | 0 | — |
case-16 | fail→fail | 17,224 | 3,852 | -78% | 1 | 1 | 0% | 3,196 | 2,084 | -35% | 0 | 0 | — |
case-17 | fail→fail | 16,404 | 2,139 | -87% | 1 | 1 | 0% | 2,911 | 1,717 | -41% | 0 | 0 | — |
case-18 | fail→fail | 16,456 | 9,582 | -42% | 1 | 1 | 0% | 3,221 | 1,899 | -41% | 0 | 0 | — |
case-19 | fail→pass | 23,398 | 10,062 | -57% | 1 | 1 | 0% | 4,118 | 2,057 | -50% | 0 | 0 | — |
case-20 | pass→pass | 7,017 | 4,638 | -34% | 1 | 1 | 0% | 1,297 | 2,156 | +66% | 0 | 0 | — |
case-21 | pass→pass | 5,266 | 4,318 | -18% | 1 | 1 | 0% | 811 | 2,003 | +147% | 0 | 0 | — |
case-22 | pass→pass | 16,705 | 11,930 | -29% | 1 | 1 | 0% | 3,163 | 3,823 | +21% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 18 counted toward the lift figure. The other 4 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +9 percentage points is the difference between those two pass rates over the 18 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.