最佳实践调研

Universal

ln-002-best-practices-researcher

by levnikolaevich

结合MCP Ref、Context7和WebSearch快速查最佳实践、官方文档与版本信息,并生成guide、manual、ADR或research文档,适合给项目补齐模式说明、API手册和架构决策记录。

415搜索与获取未扫描2026年3月5日

安装

claude skill add --url github.com/levnikolaevich/claude-code-skills/tree/master/ln-002-best-practices-researcher

文档

Paths: File paths (shared/, references/, ../ln-*) are relative to skills repo root. If not found at CWD, locate this SKILL.md directory and go up one level for repo root.

Best Practices Researcher

Research industry standards and create project documentation in one workflow.

Purpose & Scope

  • Research via MCP Ref + Context7 for standards, patterns, versions
  • Create 4 types of documents from research results:
    • Guide: Pattern documentation (Do/Don't/When table)
    • Manual: API reference (methods/params/doc links)
    • ADR: Architecture Decision Record (Q&A dialog)
    • Research: Investigation document answering specific question
  • Return document path for linking in Stories/Tasks

Phase 0: Stack Detection

Objective: Identify project stack to filter research queries and adapt output.

Detection:

IndicatorStackQuery PrefixOfficial Docs
*.csproj, *.sln.NET"C# ASP.NET Core"Microsoft docs
package.json + tsconfig.jsonNode.js"TypeScript Node.js"MDN, npm docs
requirements.txt, pyproject.tomlPython"Python"Python docs, PyPI
go.modGo"Go Golang"Go docs
Cargo.tomlRust"Rust"Rust docs
build.gradle, pom.xmlJava"Java"Oracle docs, Maven

Usage:

  • Add query_prefix to all MCP search queries
  • Link to stack-appropriate official docs

When to Use

  • ln-310-story-validator detects missing documentation
  • Need to document a pattern, library, or decision
  • Replaces: ln-321-guide-creator, ln-322-adr-creator, ln-323-manual-creator

Input Parameters

ParameterRequiredDescription
doc_typeYes"guide" / "manual" / "adr" / "research"
topicYesWhat to document (pattern name, package name, decision title, research question)
story_contextNoStory/Task context for relevance

Research Tools

ToolUse CaseQuery Pattern
ref_search_documentationStandards, patterns, RFCs"[topic] RFC standard best practices {current_year}"
context7__resolve-library-idGet library ID for docslibraryName="[topic]"
context7__query-docsLibrary API, methodstopic="[stack_prefix] [topic]"
WebSearchMarket, competitors, versions"[topic] latest {current_year}" or "[topic] vs alternatives"

Time-box: 5-10 minutes for research; skip if topic is trivial

Research Methodology by Type (for doc_type="research")

TypeFocusPrimary SourcesKey Questions
TechnicalSolution comparisonDocs, benchmarks, RFCs"Which solution fits our use-case?"
MarketIndustry landscapeReports, blogs, articles"What's the market size/trend?"
CompetitorHow others solve itProduct pages, reviews, demos"What features do competitors offer?"
RequirementsUser needsFeedback, support tickets, forums"What do customers complain about?"
FeasibilityCan we build it?PoC, prototypes, local tests"Is it technically possible?"
Feature DemandFeature viabilityCompetitor features + blogs/socials + user complaints"Is this feature worth building?"

Workflow

doc_typePurposeResearch SourceTemplateOutput PathWords
guidePattern with Do/Don't/When tableref_search (best practices)guide_template.mdguides/NN-[slug].md300-500
manualAPI/library referencecontext7__query-docsmanual_template.mdmanuals/[pkg]-[ver].md300-500
adrArchitecture decisionDialog (5 questions)adr_template.mdadrs/adr-NNN-[slug].md300-500
researchInvestigation answering questionSee Methodology table aboveresearch_template.mdresearch/rsh-NNN-[slug].md300-700

Common Workflow: Detect number (if needed) → Research → Generate from template → Validate (SCOPE, POSIX) → Save → Return path

Extract & Sections by doc_type:

  • guide: Extract principle, 2-3 do/don'ts, sources → Sections: Principle, Our Implementation, Patterns table, Sources, Related
  • manual: Extract methods, params (type/required/default), returns → Sections: Package info, Overview, Methods table, Config table, Limitations
  • adr: Dialog answers → Sections: Context, Decision, Rationale, Alternatives table, Consequences, Related
  • research: Findings by methodology → Sections: Question, Context, Methodology, Findings (tables!), Conclusions, Next Steps, Sources

Validation specifics: guide: patterns table present; manual: version in filename; adr: ISO date, status field; all: sources ≤ 1 year old

ADR Dialog (5 questions): Q1: Title? → Q2: Category (Strategic/Technical)? → Q3: Context? → Q4: Decision + Rationale? → Q5: Alternatives (2 with pros/cons)?

Output: File path for linking in Stories/Tasks; for ADR remind to reference in architecture.md; for Research suggest ADR if decision needed

Critical Rules

MANDATORY FILE CREATION:

  • ALL research MUST end with file creation (guide/manual/ADR/research document)
  • Create target directory if missing (docs/guides/, docs/manuals/, docs/adrs/, docs/research/)
  • No exceptions — file creation is required for ALL invocations

NO_CODE_EXAMPLES (ALL document types):

ForbiddenAllowed
Code snippetsTables (params, config, alternatives)
Implementation examplesASCII diagrams, Mermaid diagrams
Code blocks >1 lineMethod signatures (1 line inline)
Links to official docs

Format Priority (STRICT):

code
┌───────────────────────────────────────────────┐
│ 1. TABLES + ASCII diagrams  ←── PRIORITY      │
│    Params, Config, Alternatives, Flows        │
├───────────────────────────────────────────────┤
│ 2. LISTS (enumerations only)                  │
│    Enumeration items, file lists, tools       │
├───────────────────────────────────────────────┤
│ 3. TEXT (last resort)                         │
│    Only if cannot express as table            │
└───────────────────────────────────────────────┘
Content TypeFormat
ParametersTable: Name | Type | Required | Default
ConfigurationTable: Option | Type | Default | Description
AlternativesTable: Alt | Pros | Cons | Why Rejected
PatternsTable: Do | Don't | When
WorkflowASCII diagram: A → B → C

Other Rules:

  • Research ONCE per invocation; reuse results
  • Cite sources with versions/dates (≤ 1 year old)
  • One pattern per guide; one decision per ADR; one package per manual
  • Preserve language (EN/RU) from story_context
  • Link to stack-appropriate docs (Microsoft for .NET, MDN for JS, etc.)
  • MANDATORY: Create target directory if missing (docs/guides/, docs/manuals/, docs/adrs/, docs/research/); file creation is required

Definition of Done

  • Research completed (standards/patterns/versions extracted) - for guide/manual
  • Dialog completed (5 questions answered) - for ADR
  • Document generated with all required sections; no placeholders
  • Standards validated (SCOPE, maintenance, POSIX)
  • File saved to correct directory with proper naming
  • Path returned; README updated if placeholder present

Reference Files

  • Guide template: shared/templates/guide_template.md
  • Manual template: shared/templates/manual_template.md
  • ADR template: shared/templates/adr_template.md
  • Research template: shared/templates/research_template.md
  • Standards: docs/DOCUMENTATION_STANDARDS.md (if exists)
  • MANDATORY READ: shared/references/research_tool_fallback.md

Version: 3.0.1 Last Updated: 2026-02-14

相关 Skills

热门

股票投研9点分析框架,覆盖基本面/财务/竞品/估值/宏观/情绪等维度

搜索与获取
未扫描4.2k

SEO审计工具

by amdf01-debug

热门

搜索与获取
未扫描4.2k

智能浏览器

by ayakolin

热门

A fast Rust-based headless browser automation CLI with Node.js fallback that enables AI agents to navigate, click, type, and snapshot pages via structured commands. (Chinese localized version)

搜索与获取
未扫描4.2k

相关 MCP 服务

网页抓取

编辑精选

by Anthropic

热门

Fetch 是 MCP 官方参考服务器,让 AI 能抓取网页并转为 Markdown 格式。

这个服务器解决了 AI 直接处理网页内容时格式混乱的问题,适合需要让 Claude 分析在线文档或新闻的开发者。不过作为参考实现,它缺乏生产级的安全配置,你得自己处理反爬虫和隐私风险。

搜索与获取
84.2k

Brave 搜索

编辑精选

by Anthropic

热门

Brave Search 是让 Claude 直接调用 Brave 搜索 API 获取实时网络信息的 MCP 服务器。

如果你想让 AI 助手帮你搜索最新资讯或技术文档,这个工具能绕过传统搜索的限制,直接返回结构化数据。特别适合需要实时信息的开发者,比如查 API 更新或竞品动态。不过它依赖 Brave 的 API 配额,高频使用可能受限。

搜索与获取
84.2k

by Anthropic

热门

Puppeteer 是让 Claude 自动操作浏览器进行网页抓取和测试的 MCP 服务器。

这个服务器解决了手动编写 Puppeteer 脚本的繁琐问题,适合需要自动化网页交互的开发者,比如抓取动态内容或做端到端测试。不过,作为参考实现,它可能缺少生产级的安全防护,建议在可控环境中使用。

搜索与获取
84.2k

评论