rr-solopreneur-researcher - SKILL.md Agent Skill

name: rr-solopreneur-researcher version: 1.0.0 description: >- Role Replacement Case Study: Solopreneur Research Assistant — autonomous market research, competitor analysis, and trend scouting pipeline that replaces a dedicated research hire. Thin harness composing parallel-web-search, kb-ingest, defuddle, alphaear-search, and feynman-source-comparison into a unified research role pipeline with KB-first persistence and structured insight extraction. tags: [role-replacement, harness, research, solopreneur, market-research] triggers: - rr-solopreneur-researcher - research agent - solopreneur research - 리서치 에이전트 - 시장 조사 에이전트 - 1인 기업 리서치 - market research agent - competitor research agent do_not_use: - Full academic paper review with PM analysis (use paper-review) - HuggingFace trending intelligence (use hf-trending-intelligence) - Daily stock analysis or trading signals (use today or daily-stock-check)

Single URL content extraction without research pipeline (use defuddle directly) - General KB query without new research intent (use kb-query) composes: -.

Role Replacement: Solopreneur Research Assistant

Thin harness that replaces a dedicated research hire for solo founders and small teams by orchestrating existing research and knowledge skills into a 4-phase pipeline with KB-first persistence and structured insight extraction.

What This Replaces

Human Researcher Task	Automated By	Skill
Market landscape scanning	Multi-provider parallel web search	`parallel-web-search`
Competitor product/pricing monitoring	Finance-specific search + web extraction	`alphaear-search` + `defuddle`
Trend report compilation	Cross-source comparison with evidence matrix	`feynman-source-comparison`
Research asset archival	Markdown-first KB ingestion with YAML frontmatter	`kb-ingest`
Insight synthesis	KB compilation with cross-references	`kb-compile`
Research quality scoring	Multi-dimension rubric evaluation	`evaluation-engine`

Prerequisites

Web search available (WebSearch tool or parallel-web-search configured)
Knowledge Base topic directory exists (e.g., knowledge-bases/competitive-intel/)
No API keys required for core pipeline (web search uses built-in providers)

Architecture

Phase 1: COLLECT (parallel)
  ├── parallel-web-search (3-5 queries, multi-provider)
  ├── alphaear-search (finance/industry-specific sources)
  └── defuddle (extract clean content from discovered URLs)

Phase 2: ANALYZE (sequential)
  ├── feynman-source-comparison (cross-source agreement/disagreement matrix)
  └── evaluation-engine (score research quality: coverage, freshness, depth)

Phase 3: PERSIST (sequential)
  ├── kb-ingest (save raw sources with YAML frontmatter to KB topic)
  └── kb-compile (update wiki with new findings, cross-references)

Phase 4: SYNTHESIZE (sequential)
  ├── long-form-compressor (executive summary from full findings)
  └── Output: structured Korean research brief

Execution Modes

Mode 1: Topic Research (default)

Input: "리서치 에이전트 실행: [TOPIC]"
Output: KB-persisted research + Korean executive brief

Mode 2: Competitor Deep-Dive

Input: "경쟁사 분석: [COMPANY/PRODUCT]"
Output: Competitor profile with pricing, features, positioning, gaps

Mode 3: Trend Scout

Input: "트렌드 스캐닝: [DOMAIN]"
Output: Emerging signals ranked by evidence strength

Phase Details

Phase 1: Collect

Parse user intent to extract research topic, scope, and depth
Generate 3-5 diverse search queries (Korean + English)
Fan-out:
- parallel-web-search: broad market/industry queries
- alphaear-search: finance/pricing/market-data queries
For top-10 URLs from search results, run defuddle to extract clean markdown
Persist raw extractions to /tmp/research-{date}/ for Phase 2 input

Phase 2: Analyze

Feed all collected sources to feynman-source-comparison
- Identify agreements (consensus signals)
- Flag disagreements (conflicting data points)
- Score confidence per claim
Run evaluation-engine with research-quality rubric:
- Coverage (0-10): breadth of sources
- Freshness (0-10): recency of data
- Depth (0-10): specificity of findings
- Actionability (0-10): clarity of implications
If composite score < 6/10, loop back to Phase 1 with refined queries (max 1 retry)

Phase 3: Persist

Select KB topic based on research domain:
- Market/competitor → competitive-intel
- Technology/trend → intelligence or relevant topic
- Industry/pricing → sales-playbook or finance-policies
Run kb-ingest for each high-value source (score >= 7)
Run kb-compile to update wiki with new articles and cross-references

Phase 4: Synthesize

Compile all findings into a structured research document
Run long-form-compressor for executive summary (bullet brief format)
Output format:

## 리서치 결과: [TOPIC]

### 핵심 발견 (Executive Summary)
- [3-5 bullet points]

### 상세 분석
#### 시장 현황
#### 주요 플레이어
#### 트렌드 & 시그널
#### 기회 & 리스크

### 데이터 품질
- 소스 수: N개
- 신선도: YYYY-MM 기준
- 신뢰도 점수: X/10

### 다음 단계 제안
- [actionable next steps]

Overlap & Differentiation

Existing Agent	Overlap Area	Differentiation
`rr-market-research-analyst`	Market data, competitor analysis	This agent focuses on solopreneur-scale research without trading/quant context
`rr-knowledge-strategist`	KB persistence, wiki compilation	This agent is research-first; KS is consolidation-first
`rr-content-curator`	Web source collection	This agent analyzes and synthesizes; CC routes to Slack channels

Error Handling

Search returns < 3 results → expand query scope, add alternative keywords
defuddle fails on URL → skip and note in report, continue with other sources
KB topic doesn't exist → create minimal topic structure before ingesting
evaluation-engine score < 4 → abort with "insufficient data" report + suggested manual queries

Invocation Examples

"리서치 에이전트: AI GPU 클라우드 시장 현황 2026"
"research agent: competitor analysis for serverless inference platforms"
"시장 조사 에이전트: 한국 MSP 시장 규모 및 주요 플레이어"
"solopreneur research: pricing strategies of vLLM hosting providers"