Research Agent Team

SKILL.md

8-Agent Research Team

Overview

Simulate an 8-member specialist research team to produce deep, fact-checked, structured Markdown research reports. The pipeline runs in 5 sequential phases: scope framing → parallel research → fact verification → report writing → quality gate. Load references/agent-personas.md for detailed agent personas, thinking frameworks, and source credibility rubrics.

Pipeline

PHASE 0 — Orchestrator: Frame the Problem

Perform MECE decomposition of the research topic:

• Define 3–5 core questions the report must answer
• Assign each to the correct research axis (Market / Finance / Technology)
• Set credibility standards: prefer primary sources (filings, official stats, company IR) over secondary

Output a structured brief with 3 axes and core questions, then announce:

"PHASE 0 완료 — 리서처 3명 병렬 조사 시작"

PHASE 1 — 3 Researchers: Parallel Investigation

Run three researchers in sequence. Each uses WebSearch for real-time data, records every source URL and publication date, and flags uncertain statistics as [요확인].

Researcher A — Market/Industry

• Market size (TAM, CAGR, forecasts), competitive landscape, demand drivers, regulatory environment
• Minimum 2 WebSearch queries

Researcher B — Company/Finance

• Revenue, growth rates, margins, key partnerships, hyperscaler relationships
• Prefer: earnings calls, SEC filings, official IR pages
• Minimum 2 WebSearch queries

Researcher C — Technology/Trends

• Core tech stack, product roadmap, emerging standards, R&D investments
• Minimum 2 WebSearch queries

After all three complete, announce:

"PHASE 1 완료 — 팩트체커 교차검증 시작"

PHASE 2 — 2 Fact-Checkers: Cross-Verification

Run independently after PHASE 1:

Fact-Checker A — Number Verification

• Verify every key statistic against a primary or authoritative secondary source
• Unverifiable items: mark [미검증] with a note on the discrepancy

Fact-Checker B — Source Credibility

• Rate each source: ✅ Primary | ✅ Tier-1 Research | ⚠️ Secondary
• Flag any source that cannot be traced to a credible origin

See references/agent-personas.md for the full source credibility rubric.

Announce:

"PHASE 2 완료 — 구성작가 보고서 작성 시작"

PHASE 3 — Writer: Compose the Report

Apply the Pyramid Principle — conclusion first, then supporting evidence.

Mandatory report structure (Markdown):

1. Executive Summary — 3–5 bullet key conclusions with "so what"
2. Section 1: Market/Industry — Researcher A findings
3. Section 2: Company/Finance — Researcher B findings
4. Section 3: Technology/Trends — Researcher C findings
5. Section 4: Investment/Strategy — synthesized (include if topic warrants)
6. Risk Factors — key risks with mitigating context
7. Source Table — numbered, with credibility rating per row

Formatting rules:

• ## for sections, ### for subsections
• Tables for all multi-item comparisons (stocks, metrics, timelines)
• All [미검증] items clearly marked inline
• Investment disclaimer at the end if financial content is present

Announce:

"PHASE 3 완료 — 게이트키퍼 최종 승인 시작"

PHASE 4 — Gatekeeper: Quality Gate

Run both checklists before releasing the final report:

Logical Thinking Check

• [ ] Every conclusion is backed by at least one verified source
• [ ] No logical leap between evidence and conclusion
• [ ] Conflicting data points are acknowledged and reconciled

MECE Check

• [ ] All 3 axes (Market / Finance / Technology) are covered
• [ ] No major overlap between sections
• [ ] All PHASE 0 core questions are answered

If any item fails: return to the relevant phase, fix, then re-run the checklist. Output the final report only after all items pass.

Output Format

• Language: Match the user's language (Korean if user writes in Korean)
• Format: Markdown — headers, tables, bullet points
• Source table: Always included at the end
• Length: Proportional to topic complexity — thorough over brief

Resources

• references/agent-personas.md — Detailed personas, MECE / Logical Thinking / Pyramid Principle frameworks, source credibility rubric