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. If

shared/

is missing, fetch files via WebFetch from

https://raw.githubusercontent.com/levnikolaevich/claude-code-skills/master/skills/{path}

.

Reference Documentation Creator

Type: L2 Worker

This skill creates the reference documentation structure (docs/reference/) and smart documents (ADRs, Guides, Manuals) based on project's TECH_STACK. Documents are created only when justified (nontrivial technology choices with alternatives).

Purpose

Create the reference documentation directory structure (docs/reference/) with README hub, then generate ADRs, Guides, and Manuals only for justified (nontrivial) technology choices based on TECH_STACK from Context Store.

When to Use This Skill

This skill is a L2 WORKER invoked by ln-100-documents-pipeline orchestrator.

This skill should be used directly when:

• Creating only reference documentation structure (docs/reference/)
• Setting up directories for ADRs, guides, and manuals
• NOT creating full documentation structure (use ln-100-documents-pipeline for complete setup)

Inputs

From ln-100 (via ln-110 Context Store):

{
  "context_store": {
    "PROJECT_NAME": "my-project",
    "TECH_STACK": {
      "frontend": "React 18",
      "backend": "Express 4.18",
      "database": "PostgreSQL 15",
      "orm": "Prisma 5.0",
      "auth": "JWT",
      "cache": "Redis 7"
    },
    "DEPENDENCIES": [...],
    "flags": { "hasBackend": true, "hasDatabase": true, "hasFrontend": true }
  }
}

TECH_STACK is used for smart document creation in Phase 2.

MANDATORY READ: Load

shared/references/docs_quality_contract.md

,

shared/references/docs_quality_rules.json

, and

shared/references/markdown_read_protocol.md

.

Workflow

The skill follows a 4-phase workflow: CREATE STRUCTURE → SMART DOCUMENT CREATION → VALIDATE STRUCTURE → VALIDATE CONTENT. Phase 2 creates documents only for justified technology choices.

---

Phase 1: Create Structure

Objective: Establish reference documentation directories and README hub.

Process:

1.1 Check & create directories:

• Check if

  docs/reference/adrs/

  exists → create if missing
• Check if

  docs/reference/guides/

  exists → create if missing
• Check if

  docs/reference/manuals/

  exists → create if missing
• Check if

  docs/reference/research/

  exists → create if missing
• Log for each: "✓ Created docs/reference/[name]/" or "✓ docs/reference/[name]/ already exists"

1.2 Check & create README:

• Check if

  docs/reference/README.md

  exists
• If exists:
  • Skip creation
  • Log: "✓ docs/reference/README.md already exists, proceeding to validation"
• If NOT exists:
  • Copy template:

    ln-120-reference-docs-creator/references/reference_readme_template.md

    →

    docs/reference/README.md

  • Replace placeholders:

    • {{VERSION}}

      — "1.0.0"

    • {{DATE}}

      — current date (YYYY-MM-DD)

    • {{ADR_LIST}}

      —

      - No ADRs yet. Add the first ADR when a nontrivial decision is accepted.

    • {{GUIDE_LIST}}

      —

      - No project guides yet. Add guides when patterns need project-specific explanation.

    • {{MANUAL_LIST}}

      —

      - No package manuals yet. Add manuals only for complex external APIs.

    • {{RESEARCH_LIST}}

      —

      - No research notes yet. Add research only when a concrete question is investigated.

  • Preserve the shared opening contract and standard top sections from the template
  • Log: "✓ Created docs/reference/README.md from template"

1.3 Output:

docs/reference/
├── README.md         # Created or existing
├── adrs/            # Empty, ready for ADRs
├── guides/          # Empty, ready for guides
├── manuals/         # Empty, ready for manuals
└── research/        # Empty, ready for research documents

---

Phase 2: Smart Document Creation

Objective: Create ADRs, Guides, and Manuals for justified technology choices. Skip trivial/obvious selections.

2.1 Check Context Store:

• If no

  context_store

  provided → skip Phase 2, proceed to Phase 3
• If no

  TECH_STACK

  in context_store → skip Phase 2, proceed to Phase 3
• Log: "Context Store received with TECH_STACK: [categories count]"

2.2 Load Justification Rules:

• Read

  references/tech_justification_rules.md

• Parse category → justified/skip conditions

2.3 Analyze TECH_STACK for ADRs:

For each category in TECH_STACK (frontend, backend, database, orm, auth, cache):

1. Check if justified (from justification rules):

   • frontend

     : Justified if React/Vue/Angular/Svelte (multiple options exist)

   • backend

     : Justified if Express/Fastify/NestJS/Koa (multiple options exist)

   • database

     : Justified if PostgreSQL/MySQL/MongoDB (multiple options exist)

   • auth

     : Justified if JWT/OAuth/Session (decision required)

   • orm

     : Justified if Prisma/TypeORM/Sequelize (multiple options exist)

   • cache

     : Justified if Redis/Memcached (decision required)

2. Skip if trivial:

   • jQuery-only frontend (no framework choice)
   • Simple http server (no framework)
   • SQLite for dev only
   • No auth required
   • Raw SQL only
   • In-memory cache only

3. Create ADR if justified:

   • Determine next ADR number:

     adr-NNN-{category}.md

   • Use template:

     shared/templates/adr_template.md

   • MCP Research:

     mcp__context7__resolve-library-id(technology)

   • Fill template:
     • Title: "ADR-NNN: {Category} Selection"
     • Context: Why decision was needed
     • Decision: Technology chosen with version
     • Rationale: 3 key reasons from research
     • Alternatives: 2 other options with pros/cons
   • Save:

     docs/reference/adrs/adr-NNN-{category}.md

   • Log: "✓ Created ADR for {category}: {technology}"

4. Skip if not justified:

   • Log: "⊘ Skipped ADR for {category}: trivial choice"

2.4 Analyze TECH_STACK for Guides:

For each technology with complex configuration:

1. Check if justified:

   • Has config file with >20 lines
   • Uses custom hooks/middleware/decorators
   • Has 3+ related dependencies

2. Create Guide if justified:

   • Determine next guide number:

     NN-{technology-slug}-patterns.md

   • Use template:

     shared/templates/guide_template.md

   • MCP Research:

     mcp__Ref__ref_search_documentation("{technology} patterns {current_year}")

   • Fill template:
     • Principle: Industry standard from research
     • Our Implementation: How project uses it
     • Patterns table: 3 Do/Don't/When rows
   • Save:

     docs/reference/guides/NN-{technology}-patterns.md

   • Log: "✓ Created Guide for {technology}"

3. Skip if standard usage:

   • Log: "⊘ Skipped Guide for {technology}: standard usage"

2.5 Analyze TECH_STACK for Manuals:

For each package with complex API:

1. Check if justified:

   • Package has >10 exported methods
   • Has breaking changes in current version
   • NOT in trivial list: lodash, uuid, dotenv

2. Create Manual if justified:

   • Use template:

     shared/templates/manual_template.md

   • MCP Research:

     mcp__context7__get-library-docs(topic: "API")

   • Fill template:
     • Package info with version
     • 2-3 most used methods
     • Configuration section
   • Save:

     docs/reference/manuals/{package}-{version}.md

   • Log: "✓ Created Manual for {package}"

3. Skip if trivial API:

   • Log: "⊘ Skipped Manual for {package}: trivial API"

2.6 Report Smart Creation:

✅ Smart Document Creation complete:
  - ADRs created: [count] (justified: frontend, backend, database)
  - ADRs skipped: [count] (trivial: cache=in-memory)
  - Guides created: [count]
  - Guides skipped: [count]
  - Manuals created: [count]
  - Manuals skipped: [count]

---

Phase 3: Validate Structure

Objective: Ensure reference/README.md complies with structural requirements and auto-fix violations.

Process:

2.1 Check opening contract:

• Read

  docs/reference/README.md

  (first 5 lines)
• Check for

  <!-- SCOPE: ... -->

  tag and metadata markers
• Expected:

  <!-- SCOPE: Reference documentation hub (ADRs, Guides, Manuals) with links to subdirectories -->

• If missing:
  • Use Edit tool to add missing opening markers near the top of the file
  • Track violation:

    opening_contract_fixed = True

2.2 Check required sections:

• Load expected sections from

  references/questions.md

• Required sections:
  • "Architecture Decision Records (ADRs)"
  • "Project Guides"
  • "Package Manuals"
  • "Research"
• For each section:
  • Check if

    ## [Section Name]

    header exists
  • If missing:
    • Use Edit tool to add section with placeholder:

      ## [Section Name]
      
      {{PLACEHOLDER}}
      

    • Track violation:

      missing_sections += 1

2.3 Check Maintenance section:

• Search for

  ## Maintenance

  header
• If missing:
  • Use Edit tool to add at end of file:

    ## Maintenance
    
    **Last Updated:** [current date]
    
    **Update Triggers:**
    - New ADRs added to adrs/ directory
    - New guides added to guides/ directory
    - New manuals added to manuals/ directory
    
    **Verification:**
    - [ ] All ADR links in registry are valid
    - [ ] All guide links in registry are valid
    - [ ] All manual links in registry are valid
    - [ ] Placeholders {{ADR_LIST}}, {{GUIDE_LIST}}, {{MANUAL_LIST}} synced with files
    

  • Track violation:

    maintenance_added = True

2.4 Check POSIX file endings:

• Check if file ends with single blank line
• If missing:
  • Use Edit tool to add final newline
  • Track fix:

    posix_fixed = True

2.5 Report validation:

• Log summary:

  ✅ Structure validation complete:
    - SCOPE tag: [added/present]
    - Missing sections: [count] sections added
    - Maintenance section: [added/present]
    - POSIX endings: [fixed/compliant]
  

• If violations found: "⚠️ Auto-fixed [total] structural violations"

---

Phase 4: Validate Content

Objective: Ensure each section answers its questions with meaningful content and populate registries from auto-discovery (including documents created in Phase 2).

Process:

4.1 Load validation spec:

• Read

  references/questions.md

• Parse questions and validation heuristics for all 3 sections

4.2 Validate sections (parametric loop):

Define section parameters:

sections = [
  {
    "name": "Architecture Decision Records (ADRs)",
    "question": "Where are architecture decisions documented?",
    "directory": "docs/reference/adrs/",
    "empty_state": "No ADRs yet.",
    "glob_pattern": "docs/reference/adrs/*.md",
    "heuristics": [
      "Contains link: [ADRs](adrs/) or [adrs](adrs/)",
      "Mentions 'Architecture Decision Record' or 'ADR'",
      "Has actual list or empty-state note",
      "Length > 30 words"
    ]
  },
  {
    "name": "Project Guides",
    "question": "Where are reusable project patterns documented?",
    "directory": "docs/reference/guides/",
    "empty_state": "No project guides yet.",
    "glob_pattern": "docs/reference/guides/*.md",
    "heuristics": [
      "Contains link: [Guides](guides/) or [guides](guides/)",
      "Has actual list or empty-state note",
      "Length > 20 words"
    ]
  },
  {
    "name": "Package Manuals",
    "question": "Where are third-party package references documented?",
    "directory": "docs/reference/manuals/",
    "empty_state": "No package manuals yet.",
    "glob_pattern": "docs/reference/manuals/*.md",
    "heuristics": [
      "Contains link: [Manuals](manuals/) or [manuals](manuals/)",
      "Has actual list or empty-state note",
      "Length > 20 words"
    ]
  },
  {
    "name": "Research",
    "question": "Where are research/investigation documents stored?",
    "directory": "docs/reference/research/",
    "empty_state": "No research notes yet.",
    "glob_pattern": "docs/reference/research/*.md",
    "heuristics": [
      "Contains link: [Research](research/) or [research](research/)",
      "Has actual list or empty-state note",
      "Length > 20 words"
    ]
  }
]

For each section in sections:

1. Read section content:

   • Extract section from reference/README.md

2. Check if content answers question:

   • Apply validation heuristics
   • If ANY heuristic passes → content valid, skip to next section
   • If ALL fail → content invalid, continue

3. Auto-discovery (if content invalid or empty-state present):

   • Scan directory using Glob tool (section.glob_pattern)
   • If files found:
     • Extract filenames
     • Generate dynamic list:

       - [ADR-001: Frontend Framework](adrs/adr-001-frontend-framework.md)
       - [02-Repository Pattern](guides/02-repository-pattern.md)
       - [Axios 1.6](manuals/axios-1.6.md)
       

     • Use Edit tool to replace empty-state text with generated list
     • Track change:

       sections_populated += 1

   • If NO files:
     • Keep concise empty-state note
     • Track:

       empty_states_kept += 1

4. Skip external API calls:

   • Do NOT use MCP Ref search (template already has format examples)

4.3 Report content validation:

• Log summary:

  ✅ Content validation complete:
    - Sections checked: 4
    - Populated from auto-discovery: [count]
    - Placeholders kept (no files): [count]
    - Already valid: [count]
  

---

Complete Output Structure

docs/
└── reference/
    ├── README.md                     # Reference hub with registries
    ├── adrs/                         # Empty or with ADR files
    ├── guides/                       # Empty or with guide files
    ├── manuals/                      # Empty or with manual files
    └── research/                     # Empty or with research files

---

Reference Files

Templates

Reference README Template:

• references/reference_readme_template.md

  (v2.0.0) - Reference hub with:
  • SCOPE tags (reference documentation ONLY)
  • Three registry sections with placeholders
  • Maintenance section

Document Templates (for Phase 2 Smart Creation):

• shared/templates/adr_template.md

  - Nygard ADR format (7 sections)

• shared/templates/guide_template.md

  - Pattern documentation (Do/Don't/When)

• shared/templates/manual_template.md

  - API reference format

• shared/templates/research_template.md

  - Research/Spike documentation

Justification Rules:

• references/tech_justification_rules.md

  - Mapping: category → create/skip conditions

Validation Specification:

• references/questions.md

  (v2.0) - Question-driven validation:
  • Q1-Q3: Registry sections (ADRs, Guides, Manuals)
  • Q4-Q7: Smart document validation (ADR context, alternatives, patterns)
  • Auto-discovery hints

• MANDATORY READ: Load

  shared/references/research_tool_fallback.md

---

Best Practices

• No premature validation: Phase 1 creates structure, Phase 3 validates it
• Smart creation: Phase 2 creates documents only for justified choices
• Parametric validation: Phase 4 uses loop for 3 sections (no code duplication)
• Auto-discovery first: Scan actual files before external API calls
• Idempotent: ✅ Can run multiple times safely (checks existence before creation)
• Separation of concerns: CREATE → SMART DOCS → VALIDATE STRUCTURE → VALIDATE CONTENT

NO_CODE_EXAMPLES Rule (MANDATORY for Guides)

Guides document patterns, NOT implementations:

• FORBIDDEN: Full function implementations, code blocks > 5 lines
• ALLOWED: Do/Don't/When tables, method signatures (1 line), pseudocode (1-3 lines)
• INSTEAD OF CODE: Reference source location: "See src/hooks/usePlan.ts:15-30"
• TEMPLATE RULE: guide_template.md includes

  <!-- NO_CODE_EXAMPLES: ... -->

  tag - FOLLOW IT

Stack Adaptation Rule (MANDATORY)

• ADRs must reference stack-appropriate alternatives (Compare React vs Vue, not React vs Django)
• Guides must link to correct platform docs (Microsoft for .NET, MDN for JS)

Format Priority (MANDATORY)

Tables (Do/Don't/When) > ASCII diagrams > Lists > Text

---

Prerequisites

Invoked by: ln-100-documents-pipeline orchestrator

Requires:

• docs/

  directory (created by ln-111-root-docs-creator or already present in target project)

Creates:

• docs/reference/

  directory structure with README hub
• Validated structure and content (auto-discovery from existing files)

---

Critical Rules

• Justified creation only: Documents are created only for nontrivial technology choices with real alternatives; trivial/obvious selections are skipped
• NO_CODE_EXAMPLES: Guides document patterns (Do/Don't/When tables), not implementations; no code blocks >5 lines
• Stack adaptation: ADR alternatives must be stack-appropriate (React vs Vue, not React vs Django); links must match platform docs
• Format priority: Tables (Do/Don't/When) > ASCII diagrams > Lists > Text
• Idempotent: Checks existence before creation; safe to run multiple times

Runtime Summary Artifact

MANDATORY READ: Load

shared/references/docs_generation_summary_contract.md

Accept optional

summaryArtifactPath

.

Summary kind:

• docs-generation

Required payload semantics:

• worker = "ln-120"

• status

• created_files

• skipped_files

• quality_inputs

• validation_status

• warnings

Write the summary to the provided artifact path or return the same envelope in structured output.

Definition of Done

Before completing work, verify ALL checkpoints:

✅ Phase 1 - Structure Created:

• docs/reference/

  directory exists

• docs/reference/adrs/

  directory exists

• docs/reference/guides/

  directory exists

• docs/reference/manuals/

  directory exists

• docs/reference/research/

  directory exists

• docs/reference/README.md

  exists (created or existing)

✅ Phase 2 - Smart Documents Created (if Context Store provided):

• ADRs created for justified technology choices (nontrivial)
• ADRs skipped for trivial choices (logged)
• Guides created for technologies with complex config
• Manuals created for packages with complex API
• Each created document has real content (not placeholders)

✅ Phase 3 - Structure Validated:

• SCOPE tag and metadata markers present near top
• Quick Navigation, Agent Entry, and Maintenance sections present
• Four registry sections present (ADRs, Guides, Manuals, Research)
• Maintenance section present at end
• POSIX file endings compliant

✅ Phase 4 - Content Validated:

• All sections checked against questions.md
• Empty-state notes or actual entries present in all registries (no raw placeholders)
• No validation heuristic failures

✅ Reporting:

• Phase 1 logged: creation summary
• Phase 2 logged: smart creation (created/skipped counts)
• Phase 3 logged: structural fixes (if any)

Return Contract

Return normalized status to ln-100:

{
  "created_files": ["docs/reference/README.md", "docs/reference/adrs/adr-001-frontend.md"],
  "skipped_files": [],
  "quality_inputs": {
    "doc_paths": ["docs/reference/README.md", "docs/reference/adrs/adr-001-frontend.md"],
    "owners": {
      "docs/reference/README.md": "ln-120-reference-docs-creator",
      "docs/reference/adrs/adr-001-frontend.md": "ln-120-reference-docs-creator"
    }
  },
  "validation_status": "passed"
}

✅ Reporting (continued):

• Phase 4 logged: content updates (if any)

Meta-Analysis

MANDATORY READ: Load

shared/references/meta_analysis_protocol.md

Skill type:

documentation-creator

. Run after all phases complete. Output to chat using the

documentation-creator

format.

---

Technical Details

Standards:

• Michael Nygard's ADR Format (7 sections)
• ISO/IEC/IEEE 29148:2018 (Documentation standards)

Language: English only

---

Version: 8.2.0 Last Updated: 2025-01-12