Research Orchestrator

Diagnose your research needs, build the optimal skill chain, and guide execution step by step. No external tools — pure routing. Reads research-memory/ status to recommend what to run and when.

Purpose

Research Orchestrator is the traffic controller for all 8 research skills. It answers:

What research do I need right now?
What's already done, and what's missing?
In what order should I run the skills?
What's stale and needs refreshing?

This skill never conducts research itself. It reads research-memory/ to diagnose the current state, analyzes the user's request, and produces a step-by-step execution plan pointing to the right skills.

"Doing research without a plan is like shopping without a list — you'll miss what matters and waste time on what doesn't."

Modes

Mode	When to Use	Behavior
A: Full Research	New brand/product, empty research-memory/	Chain all 8 skills in dependency order
B: Focused Research	Specific question or area needed	Select 1-3 skills based on request
C: Refresh	Existing research is outdated	Re-run stale skills in Refresh mode
D: Validate	Strategy brief exists, need expert check	Run expert-validator only

Auto-Load Protocol

On every invocation, BEFORE any routing:

Check research-memory/ directory
If files exist → Read ALL .md files (except README.md)
For each file, assess:
- Exists? — File present with substantive content (not just scaffold headers)
- Last updated — Extract date from > Last updated: header or research-log.md
- Richness — Rich (detailed data) / Adequate / Thin (scaffold only)
- Enrichment status — For competitive-intel.md: which skills have contributed? ([competitor-finder] / [competitor-analyzer] / [competitor-visual])
- Expert validation — For strategy-brief.md: are [expert-validator] sections populated?
Read research-log.md → Extract last execution date per skill
Check brand-memory/ (read-only) → If exists, use business context to pre-fill recommendations
Build Status Dashboard (see Step 1 below)

Process

Step 1: Build Status Dashboard

Goal: Show the user exactly where their research stands.

Generate this dashboard from the Auto-Load data:

Research Status Dashboard
══════════════════════════
📊 Market Landscape    : [STATUS] [DATE_INFO]
🏢 Competitive Intel   : [STATUS] [DATE_INFO] [ENRICHMENT_INFO]
👥 Customer Insight    : [STATUS] [DATE_INFO]
💬 Customer Language   : [STATUS] [DATE_INFO]
📋 Strategy Brief      : [STATUS] [DATE_INFO] [VALIDATION_INFO]
📝 Research Log        : [LAST_ENTRY_DATE]

Status values:

✅ Complete — File exists with substantive content
⚠️ Stale (Xd ago) — Complete but outdated (30+ days)
⚠️ Partial — File exists but missing sections (e.g., competitive-intel with finder only)
❌ Missing — File doesn't exist or is empty scaffold

Staleness thresholds:

30+ days → "Update recommended"
90+ days → "Update strongly recommended"
180+ days → "Re-research needed"

Always show the dashboard first — it grounds the conversation in facts, not assumptions.

Step 2: Analyze Request + Select Mode

Goal: Match the user's request to the right mode.

Condition	Auto-Suggest
research-memory/ empty or missing + broad request	Mode A: Full Research
research-memory/ empty + specific area mentioned	Mode B: Focused (but recommend Full)
research-memory/ has data + specific area mentioned	Mode B: Focused
research-memory/ has data + "update"/"refresh"/"renew"	Mode C: Refresh
strategy-brief.md exists + "validate"/"review"/"check"	Mode D: Validate
Ambiguous request	Show dashboard → ask user to choose

Present the suggested mode with a one-line rationale. Let the user confirm or override.

Step 3: Build Execution Path

Goal: Generate the ordered list of skills to run, respecting dependencies.

Skill Dependency Map

market-scanner ─────────────────────────────────────┐
  → market-landscape.md                              │
                                                     │
competitor-finder ──┬── competitor-analyzer           │
  → competitive-    │     → enriches messaging/CTA   ├─→ research-synthesizer
     intel.md       │                                │      → strategy-brief.md
     (skeleton)     └── competitor-visual             │           │
                          → enriches design/visual    │           ▼
                                                     │    expert-validator
audience-profiler ──── voice-of-customer             │      → enriches strategy
  → customer-           → customer-language.md       │         -brief.md
     insight.md                                      │
                                                     │
                    ────────────────────────────────────┘

Group independence (for Mode A):

Group 1 (Market): market-scanner — no prerequisites
Group 2 (Competition): competitor-finder → competitor-analyzer → competitor-visual — sequential chain
Group 3 (Customer): audience-profiler → voice-of-customer — sequential chain
Group 4 (Synthesis): research-synthesizer → expert-validator — requires Groups 1+2+3

Groups 1, 2, 3 are independent of each other — user can choose the starting group.

Mode A: Full Research Path

Present the complete chain:

📋 Full Research Plan
═════════════════════
Group 1 — Market
  Step 1: market-scanner → market-landscape.md

Group 2 — Competition
  Step 2: competitor-finder → competitive-intel.md (skeleton)
  Step 3: competitor-analyzer → competitive-intel.md (+messaging)
  Step 4: competitor-visual → competitive-intel.md (+design)

Group 3 — Customer
  Step 5: audience-profiler → customer-insight.md
  Step 6: voice-of-customer → customer-language.md

Group 4 — Synthesis (after Groups 1-3)
  Step 7: research-synthesizer → strategy-brief.md
  Step 8: expert-validator → strategy-brief.md (+expert review)

💡 Groups 1-3 are independent. Which group would you like to start with?

Mode B: Focused Research Path

Map user request to target skill(s) using this table:

User Request Pattern	Target Skill	Prerequisite
Market size, trends, structure	market-scanner	None
Find competitors, competitive set	competitor-finder	None
Competitor website, messaging, pricing	competitor-analyzer	competitor-finder ✅
Competitor design, visual, screenshots	competitor-visual	competitor-finder ✅
Target audience, segments, journey	audience-profiler	None
Customer language, reviews, community	voice-of-customer	audience-profiler recommended
Strategy brief, cross-analysis, synthesis	research-synthesizer	market-scanner + competitor-finder + audience-profiler ✅
Expert review, validate strategy	expert-validator	research-synthesizer ✅

Check prerequisites against dashboard status
If prerequisite missing → include it in the path
Present the focused path

Mode C: Refresh Path

Read research-log.md → calculate days since last run per skill
Identify stale skills (30/90/180-day thresholds)
Build refresh path — stale skills only, in dependency order
Cascade rule: If a data-producing skill refreshes, downstream synthesis may need refresh too:
- market-landscape refreshed → flag research-synthesizer for refresh
- competitive-intel refreshed → flag research-synthesizer for refresh
- customer-insight refreshed → flag voice-of-customer + research-synthesizer for refresh

Mode D: Validate Path

Check strategy-brief.md exists with content
If YES → direct to expert-validator
If NO → explain prerequisite chain: research-synthesizer first (and its prerequisites if missing)

Step 4: Guide Execution

Goal: Hand off to the first skill and provide ongoing navigation.

Language: 사용자가 언어를 지정하면 대시보드 및 안내 텍스트를 해당 언어로 출력합니다. 개별 스킬 호출 시 동일한 언어 설정을 전달합니다.

Initial handoff:

▶ Next skill: [SKILL_NAME]
  Purpose: [one-line description]
  Input needed: [what the user should prepare]

Say "[SKILL_NAME] 실행해줘" to start.

After each skill completes:

Provide a transition prompt:

✅ [COMPLETED_SKILL] done → [output_file] saved

▶ Next: [NEXT_SKILL_NAME]
  Purpose: [one-line description]

Continue? Say "[NEXT_SKILL_NAME] 실행해줘"

After all research skills complete:

If strategy-brief.md exists with Next Steps → bridge to marketing execution:

🎯 Research complete! Your strategy brief recommends:
  1. [Next Step 1] → [marketing skill name]
  2. [Next Step 2] → [marketing skill name]
  3. [Next Step 3] → [marketing skill name]

Ready to execute? Pick a next step to start.

Skill Quick-Reference

Skill	MCP Tool	Output	Time Est.
market-scanner	Perplexity	market-landscape.md	5-10 min
competitor-finder	Perplexity	competitive-intel.md (skeleton)	5-10 min
competitor-analyzer	Firecrawl	competitive-intel.md (+messaging)	10-15 min
competitor-visual	Playwright	competitive-intel.md (+design)	10-15 min
audience-profiler	Perplexity	customer-insight.md	5-10 min
voice-of-customer	Perplexity	customer-language.md	5-10 min
research-synthesizer	None	strategy-brief.md	5-10 min
expert-validator	Task Agents	strategy-brief.md (+expert)	10-15 min

Quality Checklist

Before presenting a research plan, verify:

Status Dashboard shown with accurate file states and dates
Mode selection has clear rationale tied to dashboard state + user request
Execution path respects all skill dependencies (no skill runs before its prerequisite)
Each step in the path names the skill, its output, and what the user needs to provide
Focused mode includes missing prerequisites in the path
Refresh mode uses staleness thresholds consistently (30/90/180 days)
Refresh cascade logic applied (upstream refresh → downstream re-synthesis flagged)
Post-completion bridge to marketing execution skills (when strategy-brief.md has Next Steps)

Example: Full Research (Abbreviated)

User: "우리 브랜드 리서치를 처음부터 해야 하는데"

Dashboard: 📊 Market Landscape: ❌ Missing · 🏢 Competitive Intel: ❌ Missing · 👥 Customer Insight: ❌ Missing · 💬 Customer Language: ❌ Missing · 📋 Strategy Brief: ❌ Missing

Mode: A — Full Research (research-memory/ is empty)

Plan: market-scanner → competitor-finder → competitor-analyzer → competitor-visual → audience-profiler → voice-of-customer → research-synthesizer → expert-validator

Next: market-scanner — "비즈니스/제품 설명을 알려주시면 시작합니다."

Example: Focused Research (Abbreviated)

User: "경쟁사 웹사이트 메시징을 좀 더 깊게 분석해줘"

Dashboard: 📊 Market Landscape: ✅ Complete (2025-01-15) · 🏢 Competitive Intel: ✅ Complete (2025-01-12) · ...

Mode: B — Focused (competitor-analyzer targets website messaging) Prerequisite: competitor-finder ✅ already complete

Plan: competitor-analyzer only Post-completion: "competitive-intel.md가 업데이트되었습니다. research-synthesizer로 전략 브리프도 갱신할까요?"

Example: Refresh (Abbreviated)

User: "3개월 전 리서치인데 업데이트 필요해"

Dashboard: 📊 Market Landscape: ⚠️ Stale (95d) · 🏢 Competitive Intel: ⚠️ Stale (92d) · 👥 Customer Insight: ⚠️ Stale (90d) · 💬 Customer Language: ✅ 45d · 📋 Strategy Brief: ⚠️ Stale (88d)

Mode: C — Refresh (3 files exceed 90-day threshold) Plan: market-scanner (Refresh) → competitor-finder (Refresh) → audience-profiler (Refresh) → research-synthesizer (Refresh) Cascade: Strategy brief refresh needed because 3 upstream sources refreshed.

What This Skill Does NOT Do

Conduct research → Use the individual research skills (market-scanner, competitor-finder, etc.)
Write to research-memory/ → Each skill writes its own output; this skill only reads
Execute marketing → Use execution skills (brand-voice, copy, SEO, email, etc.)
Replace skill selection judgment → It recommends; the user decides

Research Orchestrator stays focused on routing — diagnosing what's needed, building the path, and guiding execution.

research-orchestrator