education-tutor-research

Tutor Research

Overview

Takes a learning question and investigates multiple paths in parallelto synthesize reliable, up-to-date learning evidence. It runs two axes simultaneously: ① Official documentation axis — use WebSearch to find the official documentation URLs for libraries, SDKs, frameworks, and CLIs, then use WebFetch to verify the body text; ② Latest information axis — use WebSearch to investigate the latest changes, tutorials, concept comparisons, and real-world cases. Cross-verify sources to block hallucinations, and education-learning-materialproduce a structured synthesis that can be used immediately.

Trigger keywords

investigate, research, latest information, official documentation, parallel investigation, accurate content, tell me, compare, difference, how it works

Workflow

Step 1: Analyze the question → derive research axes

Decompose the question into sub-axes for answering it.

• Official documentation axis: Does a specific library, SDK, tool, or API appear? (e.g., React 19, Next.js, claude code CLI, FastAPI) → derive official documentation site/URL candidates
• Latest information axis (WebSearch target): concepts, trends, comparisons, "when to use what", recent changes → 2-4 search queries
• Codebase axis (optional): when including code inside the user's project as a learning target — if the environment supports subagents (Agent(Explore)), use it; otherwise read the files directly to investigate

Step 2: Execute parallel research

[HARD] Independent investigations are in parallel in one turn executed (serial is prohibited). Call the following simultaneously.

• Official documentation: use WebSearch to find the official documentation URL → use WebFetch to verify the body of that page
• WebSearch: 2-4 derived queries (latest changes, comparisons, tutorials)
• (For broad topics, if the environment supports subagents) Agent(Explore)/Agent(general-purpose) researchers for each sub-axis parallel fan-out. In environments that do not support subagents (e.g., Claude Cowork Desktop), run WebSearch directly in parallel per axis.

For broad, multi-faceted topics (where source cross-verification is key) and the environment supports /deep-research, you can propose the bundled /deep-research <질문> workflow (first inform the user that the token cost is greater than a single search). If it is not supported, cross-verify directly with multiple WebSearch queries.

[한 턴 안에서 병렬 호출]
  ├─ WebSearch: "<라이브러리> 공식 문서 <핵심 토픽>" → WebFetch: <공식 문서 URL>
  ├─ WebSearch: "<주제> 2026 최신"
  ├─ WebSearch: "<개념 A> vs <개념 B> 차이 언제 사용"
  └─ (선택, 서브에이전트 지원 시) Agent(Explore): 사용자 프로젝트 내 관련 코드 패턴 — 미지원 시 직접 파일 읽기

GLM backend note: moai glm or moai cgin the GLM pane, WebSearch and WebFetch are routed to z.ai MCP tools. .claude/rules/moai/core/glm-web-tooling.mdFollow the routing table in

Step 3: Cross-verification + synthesis

• Source verification: Only cite URLs whose existence and content have been verified via WebFetch from WebSearch results. Do not generate URLs that are not in the search results (anti-hallucination policy).
• Conflict handling: If official documentation and web sources differ, prioritize official documentation (official site), and mark web sources as supplementary/context. For conflicts due to version differences, specify the version.
• Level adaptation: Adjust terminology difficulty to match the learner level of education-learning-project (for beginners, start from prerequisite concepts).

Step 4: Output the synthesis (education-learning-material input specification)

education-learning-materialCreate structured Markdown that can be consumed directly.

## 리서치 종합: <주제>
- 학습자 수준: <입문/초급/중급/고급>

### 핵심 개념 (요약)
- 개념 1 — 한 줄 정의 + 근거 출처
- ...

### 도식 후보 (mermaid 소스 초안)
- 개념 관계도 / 플로우 / 시퀀스 중 적합한 것 1-2개 초안

### 예제·코드 (출처·버전 표기)
- 공식 문서 기준 최소 예제

### 자주 헷갈리는 점 / 비교
- A vs B, 언제 무엇을 쓰나

### 출처
- 공식 문서: <공식 사이트 URL·버전>
- 웹: <검증된 URL 목록>

Deliverables

• Structured research synthesis (Markdown, with sources)
• Mermaid diagram source draft (passed to education-learning-material)
• Verified source list (official documentation + web URLs)

Usage examples

Example 1: "Research the latest information on how claude code subagents work" → WebSearch+WebFetch (claude code official documentation) + WebSearch (latest changes) in parallel → synthesis + sequence diagram draft

Example 2: "Prepare an explanation of React 19 use hooks based on official documentation" → Centered on WebSearch+WebFetch (React 19 official documentation) + WebSearch (practical examples) → synthesis of examples and comparisons

Example 3: "Research the accurate rules and example sentences for the English past perfect subjunctive" → WebSearch (grammar rules and example sentences) in parallel → synthesis of a rules table + example sentences (library-independent → omit the official documentation axis)

Precautions

• No assertions without sources: Mark unverified content as "needs verification".
• A Sources section is mandatory when using WebSearch — include only verified URLs as Markdown links.
• Official documentation lookup: For official documentation, confirm the URL via WebSearch, then retrieve the body directly via WebFetch (do not generate URLs that are not in the search results).
• Do not ask the user directly: If additional information is needed, leave it as a "needs verification" item in the synthesis, and the orchestrator handles decisions via AskUserQuestion.

Related skills

• moai-tutor:education-learning-project: The milestone key question becomes the input for this skill
• moai-tutor:education-learning-material: Convert this skill's synthesis into diagrams and HTML learning materials

When not to use this skill

• Literature review and citation management for academic papers → moai-tutor:education-research-assistant
• Market and business research → moai-consultant:consult-market
• Deep multi-source cross-verification reports (single question, non-learning) → If the environment supports it, bundled /deep-research workflow, otherwise direct research via multiple WebSearch queries