convez.ai
SkillBusca & PesquisaAvançado
LawlietLuoJia

Business Deep Research para Claude Code

LawlietLuoJia/claude-code-business-deep-research

Pesquisa empresarial estruturada com metodologia consulting e análise baseada em frameworks

Ver no GitHub
InicialPluginScore · de 0 a 100

Skill estruturada para pesquisa aprofundada em questões empresariais, utilizando metodologia inspirada em consulting com múltiplas rodadas de busca, análise baseada em frameworks e relatórios estruturados com fontes citadas. Produz documentos compartilháveis (brief, log de buscas, relatório e gaps) em vez de respostas superficiais.

Instalação

$npx convez-ai LawlietLuoJia/claude-code-business-deep-research

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
2

Casos de uso

  • Análise de entrada em novos mercados com evidência estruturada
  • Avaliação competitiva e análise de moats com cross-referencing de fontes
  • Relatórios de decisão com opções e recomendações fundamentadas
  • Identificação de gaps de pesquisa antes de investigação primária

Veredito

Ferramenta sólida para elevar a qualidade de pesquisas empresariais em Claude Code, indo além de buscas superficiais. Exige clareza na formulação de questões e beneficia quem precisa de análises estruturadas e rastreáveis para tomada de decisão.

README

A tradução para o português deste README ainda está na fila.

Business Deep Research — Claude Code Skill

中文文档

A structured deep-research skill for Claude Code that guides LLM-powered research through a consulting-style methodology, producing structured reports with cited sources and acknowledged gaps.

Background

When you need to research a business question — entering a new market, evaluating a competitor, assessing a technology trend — a single LLM search gives you a surface-level answer. The information might be useful, but it's not sufficient for decision-making.

The gaps are structural:

  • No methodology: The LLM searches and summarizes, but doesn't follow a systematic process
  • No cross-referencing: Claims from one source aren't verified against others
  • No gap tracking: The output doesn't tell you what it couldn't find
  • No structured output: You get a wall of text, not documents you can share with stakeholders

Business Deep Research addresses these by embedding a research methodology into the skill itself.

How It Works

The skill guides the LLM through a structured process modeled on management consulting research:

Step 1: Scope the Research

Before searching, the skill helps define:

  • The precise research question
  • The depth level needed (quick scan vs. full deep-dive)
  • Which research framework fits the question shape
  • What deliverables to produce

Step 2: Multi-Round Retrieval

Instead of one search, the skill runs multiple retrieval rounds. Each round has a quality gate:

  • After round 1: Is the source pool sufficient to proceed?
  • After round 2: Does the evidence support the chosen framework's requirements?
  • After round 3: Are gaps identified before moving to synthesis?

If a gate fails, the skill adjusts queries or flags limitations honestly.

Step 3: Framework-Driven Analysis

The skill selects one of five research frameworks based on the question:

Question ShapeFrameworkWhen to Use
How did X evolve to its current state?Evolution LensUnderstanding history, trajectory, path dependencies
How does X compare to peers?Competitive LensPeer analysis, moat evaluation, positioning
What's the structure of X's market?Market LensMarket sizing, value chain, regulation, risk
What should we do about X?Consulting FrameworkDecision memo, options analysis, recommendation
What patterns exist in qualitative data?Research SynthesisInterview/survey/feedback analysis

The framework determines what evidence to prioritize and how to structure the analysis. A second framework can be added if it addresses a distinct sub-question.

Step 4: Structured Output

The skill produces four documents, not a single chat response:

DocumentContent
Research BriefQuestion, scope, depth level, chosen framework, constraints
Search LogEvery query run, every source found, inclusion/exclusion rationale
ReportFramework-driven analysis with inline citations to sources
Assumptions & GapsWhat assumptions were made, what couldn't be verified, what needs primary research

The "Assumptions & Gaps" section is particularly important: it tells you what the research doesn't know, so you can make informed decisions about what to trust and what to verify independently.

Step 5: Review

A quality check against the evaluation rubric before finalizing.

Design Decisions

Why framework-driven? Without a framework, research tends to be unfocused — gathering information without a clear analytical structure. The framework keeps the research targeted and the output organized.

Why multiple retrieval rounds? One search rarely surfaces sufficient evidence. Multiple rounds allow the skill to refine queries based on what it found (or didn't find) previously.

Why explicit gap acknowledgment? LLMs tend to present findings with false confidence. Forcing the skill to document what it doesn't know counteracts this tendency and gives the reader honest context.

Why structured artifacts? Research output often needs to be shared with stakeholders who weren't in the conversation. Structured documents are more useful than chat logs.

Free toolchain? The skill uses whatever search tools are available in the Claude Code environment — no paid API dependencies. It adapts its retrieval strategy accordingly.

Installation

# Project-level (recommended)
cp -r . /your-project/.claude/skills/business-deep-research/

# Global
cp -r . ~/.claude/skills/business-deep-research/

Usage

Trigger phrases:

深度研究 / 商业研究 / 竞品分析 / 市场研究 / 课题研究 / 研究报告
deep research / competitive research / market analysis / business decision memo

Project Structure

business-deep-research/
├── SKILL.md                          # Skill definition
├── agents/openai.yaml                # Agent configuration
├── assets/templates/                 # 4 output templates
│   ├── research-brief-template.md
│   ├── search-log-template.md
│   ├── report-template.md
│   └── assumptions-and-gaps-template.md
├── evals/                            # Evaluation criteria
│   ├── evals.json
│   ├── evaluation-rubric.md
│   └── baseline-pressure-notes.md
├── references/                       # 14 reference documents
│   ├── framework-selector.md         # Framework routing logic
│   ├── competitive-lens.md
│   ├── market-lens.md
│   ├── evolution-lens.md
│   ├── consulting-problem-solving.md
│   ├── research-synthesis-lens.md
│   ├── search-quality.md
│   ├── strict-verification.md
│   ├── retrieval-rounds.md
│   ├── writing-style-cn.md           # Chinese writing quality
│   ├── depth-profile.md
│   ├── output-contract.md
│   ├── runtime-adapters.md
│   └── long-report.md
└── schemas/                          # 4 JSON schemas
    ├── source.schema.json
    ├── evidence.schema.json
    ├── claim.schema.json
    └── run-manifest.schema.json

License

MIT

Ver o original no GitHub

Bom para

Relacionados