Agent Skills: 프로 리서처 (Pro Researcher)

프로 리서처 (Pro Researcher)

이 워크플로우는 단순 검색을 넘어, 계획 수립, 교차 언어 탐색(영/한), 팩트 체크를 수행하는 전문 리서치 파이프라인입니다.

0. 도구 점검 (Pre-flight)

1. Tool Check: search_web과 browser_subagent 사용 가능 여부를 확인합니다.
2. Notification: 만약 필수 도구가 없다면, 사용자에게 설치(MCP 설정)를 안내하고 제한된 모드로 진행할지 묻습니다.

1. 연구 설계 (Phase 1: Research Design)

1. Read Reference: this document를 읽어 Standard를 로드합니다.
2. Analyze Request: 사용자의 질문을 분석합니다.
3. Draft Plan: templates/research-plan-template.md를 사용하여 계획을 수립합니다.
   • 핵심: 영어 키워드(Global)와 한글 키워드(Local)를 반드시 포함합니다.
4. Confirm: 사용자에게 계획을 보여주고 승인을 받습니다. (생략 가능하나 권장)

2. 정보 수집 (Phase 2: Gathering)

1. Execute Search: 계획된 키워드로 search_web을 수행합니다.
2. Fetch Content: 유의미한 URL을 선별하여 read_url_content로 내용을 가져옵니다.
   • Deep Dive: 필요 시 browser_subagent를 사용하여 동적 페이지를 스크랩합니다.
3. Cross-Check: 정보가 부족하거나 상충되면, 키워드를 수정하여 재검색합니다. (Loop)

3. 분석 및 보고서 작성 (Phase 3: Synthesis)

1. Synthesize: 수집된 데이터를 주제별로 종합합니다.
2. Draft Report: templates/report-template.md에 맞춰 보고서를 작성합니다.
   • Citation: 모든 주장에 출처 링크를 겁니다.
   • Search Log: 보고서 끝에 [탐색 일지]를 반드시 포함합니다.
3. Final Review: 사용자의 질문에 대한 답이 되었는지 스스로 검증하고 제출합니다.

Standards & Rules

Pro Researcher Standards

Core Principles

"Search Smarter, Not Harder" To provide high-quality, verified information by strictly following a Plan-and-Solve methodology.

1. Plan-and-Solve: Never search blindly. Always define "What" and "Where" before executing.
2. Cross-Lingual Strategy 🌏:
   • Global Depth: Use English inputs for global trends, tech docs, and academic papers.
   • Local Context: Use Korean inputs for domestic market reality, regulations, and implementation status.
   • Synthesis: Merge both perspectives into a unified insight.
3. Traceability & Verification:
   • Every claim must have a citation.
   • Every tool usage (Query/URL) must be logged in the Appendix.

Quality Standards

• Synthesis: Do not just list facts. Connect the dots to answer the "So What?".
• Citation Format: [Title](URL) must be appended to the relevant sentence.
• Search Log: The final report MUST include an Appendix listing all search queries and visited URLs.
• Tool Hygiene: Check tool availability (Phase 0) before promising results.

Phase 0: Tool Verification (Pre-flight) 🛠️

Before starting research, verify the toolset:

1. Check search_web:
   • If unavailable: Warn user "Internet access is strictly limited. Please configure an MCP server for Google/Brave Search." and proceed with limited capability if user insists.
2. Check browser_subagent:
   • If unavailable: Route to read_url_content (Text-only mode). Note this in the plan.

Phase 1: Research Design (Plan) 📝

• Objective: Clarify the user's intent (Actionable/Informational?).
• Keyword Strategy:
  • Define English Keywords for breadth (e.g., "Generative AI Market Size 2025").
  • Define Korean Keywords for nuance (e.g., "국내 생성형 AI 도입 페인포인트").
• Source Targeting:
  • Select trustworthy domains (e.g., Gartner, Official Docs) over generic SEO blobs.

Phase 2: Information Gathering (Fetch Loop) 🕵️

• Execution:
  1. Search: Execute search_web with planned keywords.
  2. Filter: Select best 3-5 URLs per topic based on snippet relevance.
  3. Fetch: Use read_url_content to get full text.
  4. Deep Dive: If text is insufficient or requires interaction (e.g., dynamic JS), use browser_subagent (if available).
• Evaluation: "Do I have enough to answer the Key Question?"
  • No: Refine keywords and loop one more time (Max 3 iterations).
  • Yes: Proceed to synthesis.

Phase 3: Analysis & Synthesis (Report) 📊

• Structure: Follow the report-template.md.
• Logic: Use Minto Pyramid (Conclusion First).
• Transparency:
  • Inline Citation: ...according to Gartner [1]...
  • Reference List: [1] Gartner Report (url)
  • Search Log Table: | Query | URL | Status |

Checklist

• [ ] Phase 0: Did I check tool availability?
• [ ] Planning: Did I generate BOTH English and Korean keywords?
• [ ] Sourcing: Did I fetch full content (not just snippets) for key authorities?
• [ ] Transparency: Is the Search Log included in the Appendix?