Paths: File paths (references/, ../ln-*) are relative to this skill directory.

Project Documentation Coordinator

Type: L2 Coordinator Category: 1XX Documentation Pipeline

Runtime-backed docs coordinator. The runtime owns context assembly checkpoints, conditional worker fan-out, and docs-generation summary aggregation.

Purpose & Scope

• Single context gathering — analyzes project once, builds Context Store
• Project type detection — determines hasBackend, hasDatabase, hasFrontend, hasDocker
• Delegates to 5 workers — passes Context Store to each worker
• Aggregates results — collects status from all workers, returns summary
• Solves the "context loss" problem by gathering data once and passing explicitly
• Builds the routing table used by ln-100 docs-quality repair loop

Runtime Contract

MANDATORY READ: Load references/coordinator_runtime_contract.md, references/docs_runtime_contract.md, references/docs_generation_summary_contract.md

Runtime family: docs-runtime

Identifier:

• project-docs

Phases:

1. PHASE_0_CONFIG
2. PHASE_1_CONTEXT_ASSEMBLY
3. PHASE_2_DETECTION
4. PHASE_3_DELEGATE
5. PHASE_4_AGGREGATE
6. PHASE_5_SELF_CHECK

Worker summary contract:

• ln-111..115 may receive summaryArtifactPath
• each worker writes or returns docs-generation summary envelope
• ln-110 consumes worker summaries, not free-text worker prose

Invocation (who/when)

• ln-100-documents-pipeline: Invoked as first L2 coordinator in documentation pipeline
• Never called directly by users

Inputs

From ln-100 (optional source cleanup phase):

{
  "SOURCE_DOC_NOTES": {
    "architecture_notes": { "sections": [...], "diagrams": [...] },
    "requirements_notes": { "functional": [...] },
    "principles_notes": { "principles": [...], "anti_patterns": [...] },
    "tech_stack_notes": { "frontend": "...", "backend": "...", "versions": {} },
    "api_notes": { "endpoints": [...], "authentication": "..." },
    "database_notes": { "tables": [...], "relationships": [...] },
    "runbook_notes": { "prerequisites": [...], "install_steps": [...], "env_vars": [...] },
    "infrastructure_notes": { "servers": [...], "domains": [...], "ports": {} }
  }
}

SOURCE_DOC_NOTES enriches the standard context store before worker dispatch. Workers consume normalized context fields only; there is no legacy-specific override branch.

MANDATORY READ: Load references/docs_quality_contract.md.

Architecture

ln-110-project-docs-coordinator (this skill)
├── Phase 1: Context Gathering (ONE TIME)
├── Phase 2: Delegate to Workers
│   ├── ln-111-root-docs-creator → 4 root docs (ALWAYS)
│   ├── ln-112-project-core-creator → 3 core docs (ALWAYS)
│   ├── ln-113-backend-docs-creator → 2 docs (if hasBackend/hasDatabase)
│   ├── ln-114-frontend-docs-creator → 1 doc (if hasFrontend)
│   └── ln-115-devops-docs-creator → 2 docs (1 always + 1 if hasDocker)
└── Phase 3: Aggregate Results

Workflow

Phase 1: Context Gathering (ONE TIME)

1.1 Auto-Discovery (scan project files):

1.2 Detect Project Type:

1.3 User Materials Request:

• Ask: "Do you have existing materials (requirements, designs, docs)?"
• If provided: Extract answers for Context Store

1.4 MCP Research (for detected technologies):

• Use Context7/Ref for best practices
• Store in Context Store for workers

1.5 Build Context Store:

{
  "PROJECT_NAME": "my-project",
  "PROJECT_DESCRIPTION": "...",
  "TECH_STACK": { "frontend": "React 18", "backend": "Express 4.18", "database": "PostgreSQL 15" },
  "DEPENDENCIES": [...],
  "SRC_STRUCTURE": { "controllers": [...], "services": [...] },
  "ENV_VARIABLES": ["DATABASE_URL", "JWT_SECRET"],
  "DEV_COMMANDS": { "dev": "npm run dev", "test": "npm test" },
  "DOCKER_SERVICES": ["app", "db"],
  "DEPLOYMENT_SCALE": "single",
  "DEVOPS_CONTACTS": [],
  "HAS_GPU": false,
  "SERVER_INVENTORY": [],
  "DOMAIN_DNS": [],
  "ARTIFACT_REPOSITORY": {},
  "HOST_REQUIREMENTS": {},
  "flags": { "hasBackend": true, "hasDatabase": true, "hasFrontend": true, "hasDocker": true }
}

DEPLOYMENT_SCALE detection rules:

• "single" (default): No deploy.replicas, no scaling keywords in README
• "multi": deploy.replicas > 1 OR load balancer mentioned
• "auto-scaling": auto-scaling keywords in README/docker-compose
• "gpu-based": runtime: nvidia in docker-compose

DEVOPS_CONTACTS fallback chain:

1. CODEOWNERS file → extract maintainers
2. package.json author/contributors → extract names/emails
3. git log → top 3 frequent committers
4. If all empty → [TBD: Provide DevOps team contacts]

1.6 Merge Normalized Source Notes (if provided by ln-100):

• Check if SOURCE_DOC_NOTES was passed from ln-100 cleanup phase
• If present, merge supporting facts into the standard Context Store:

  contextStore.SOURCE_DOC_NOTES = input.SOURCE_DOC_NOTES
  

• Enrichment intent:
  • architecture notes strengthen architecture evidence
  • requirement notes strengthen requirements evidence
  • tech-stack notes clarify versions and rationale
  • principles, API, database, runbook, and infrastructure notes provide supplemental factual input
• Workers always generate from normalized context + current project sources
• If no SOURCE_DOC_NOTES are present: rely on direct auto-discovery + current canonical docs

Phase 2: Delegate to Workers

MANDATORY: All applicable workers MUST be invoked. Workers run in parallel via Agent tool for context isolation.

2.1 Always invoke (parallel):

• ln-111-root-docs-creator with Context Store
• ln-112-project-core-creator with full Context Store
• ln-115-devops-docs-creator with Context Store (infrastructure.md always; runbook.md internally conditional on hasDocker)

2.2 Conditionally invoke:

• ln-113-backend-docs-creator if hasBackend OR hasDatabase
• ln-114-frontend-docs-creator if hasFrontend

Invocation (parallel via Agent tool):

Agent(description: "{doc_type} docs via {worker}",
     prompt: "Invoke Skill(skill: \"{worker}\") with context below.\n\nCONTEXT: {contextStore}",
     subagent_type: "general-purpose")

Delegation Rules:

• Pass Context Store and flags to workers via Agent+Skill pattern
• Wait for all Agent completions
• Collect normalized result (created_files, skipped_files, quality_inputs, validation_status)

Phase 3: Aggregate Results

1. Col