ln-100-documents-pipeline（100 个文档管道）

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}.

Type: L1 Top Orchestrator Category: 1XX Documentation Pipeline

Documentation Pipeline (Orchestrator)

This skill orchestrates the creation of a complete documentation system by invoking one L2 coordinator plus 3 L2 workers. The coordinator (ln-110) delegates to 5 L3 workers for project docs; other L2 workers handle reference, tasks, and test domains. Each component validates its own output, and the orchestrator owns the final docs-quality acceptance gate.

Purpose

Top-level orchestrator that creates a complete project documentation system in one invocation. Chains ln-110 coordinator + ln-120/130/140 workers sequentially, then runs global cleanup (deduplication, orphan reporting, cross-link validation).

Runtime Contract

MANDATORY READ: Load shared/references/coordinator_runtime_contract.md, shared/references/docs_pipeline_runtime_contract.md, shared/references/docs_generation_summary_contract.md

Runtime family: docs-pipeline-runtime

Identifier:

• docs-pipeline

Phases:

1. PHASE_0_CONFIG
2. PHASE_1_LEGACY_SCAN
3. PHASE_2_CONFIRMATION
4. PHASE_3_DELEGATE
5. PHASE_4_QUALITY_GATE
6. PHASE_5_CLEANUP
7. PHASE_6_SELF_CHECK

Component summary contract:

• ln-110, ln-120, ln-130, ln-140 write or return docs-generation envelopes
• ln-100 consumes those summaries plus docs-quality checkpoints as orchestration SSOT

Architecture

ln-100-documents-pipeline (L1 Top Orchestrator - this skill)
|- ln-110-project-docs-coordinator (L2 Coordinator)
|  |- ln-111-root-docs-creator (L3 Worker) -> 5 docs
|  |- ln-112-project-core-creator (L3 Worker) -> 3 docs
|  |- ln-113-backend-docs-creator (L3 Worker) -> 2 conditional
|  |- ln-114-frontend-docs-creator (L3 Worker) -> 1 conditional
|  `- ln-115-devops-docs-creator (L3 Worker) -> 1 always + 1 conditional
|- ln-120-reference-docs-creator (L2 Worker)
|- ln-130-tasks-docs-creator (L2 Worker)
`- ln-140-test-docs-creator (L2 Worker - optional)

When to Use This Skill

This skill should be used when:

• Start a new IT project and need complete documentation system at once
• Use automated workflow instead of manually invoking multiple workers
• Create entire documentation structure (AGENTS.md canonical + CLAUDE.md import stub -> docs/) in one go
• Prefer orchestrated CREATE path over manual skill chaining
• Need automatic global cleanup (deduplication, orphaned files, consolidation)

Alternative: If you prefer granular control, invoke workers manually:

1. ln-110-project-docs-coordinator - Root + Project docs (coordinates 5 L3 workers)
2. ln-120-reference-docs-creator - reference/ structure
3. ln-130-tasks-docs-creator - tasks/README.md + kanban
4. ln-140-test-docs-creator - tests/README.md (optional)

MANDATORY READ: Load shared/references/docs_quality_contract.md.

Note: Each worker validates its own local output in Phase 2/3. The orchestrator still owns the final docs-quality manifest, verification, and repair loop before cleanup can continue.

Workflow

The skill follows a 5-phase orchestration workflow: Legacy Migration (optional) -> User confirmation -> Invoke coordinator + workers sequentially -> Global cleanup -> Summary.

Phase 0: Source Documentation Cleanup (OPTIONAL)

Objective: Detect non-canonical documentation, normalize any useful facts, and remove duplicate documentation paths before generation.

Trigger: Runs at pipeline start when the repo already contains documentation outside the canonical layout.

Process:

0a. Non-Canonical Source Detection:

• Scan project for non-standard documentation using patterns from references/legacy_detection_patterns.md:

- Root.md files: ARCHITECTURE.md, REQUIREMENTS.md, STACK.md, API.md, DATABASE.md, DEPLOYMENT.md - Non-canonical folders: documentation/, doc/, wiki/, or docs/ with wrong structure - README.md sections: ## Architecture, ## Tech Stack, ## Requirements, etc. - CONTRIBUTING.md sections: ## Development, ## Code Style, ## Coding Standards

• Build source_manifest: list of {path, detected_type, target_doc, confidence}
• If no non-canonical docs are found -> skip to Phase 1

0b. Fact Extraction:

• For each detected source file:

- Parse markdown structure (headers, lists, code blocks) - Apply type-specific extractor (MANDATORY READ: Load references/legacy_detection_patterns.md): - architecture_extractor -> {layers[], components[], diagrams[]} - requirements_extractor -> {functional[], non_functional[]} - tech_stack_extractor -> {frontend, backend, database, versions} - principles_extractor -> {principles[], anti_patterns[]} - api_spec_extractor -> {endpoints[], authentication} - database_schema_extractor -> {tables[], relationships[]} - runbook_extractor -> {prerequisites[], install_steps[], env_vars[]} - infrastructure_extractor -> {servers[], domains[], ports[], services[], artifacts{}, cicd{}} - Score content quality (0.0-1.0)

• Store results in source_notes

0c. User Confirmation:

• Display detected non-canonical files: Non-Canonical Documentation Detected: | File | Type | Confidence | Target | |------|------|------------|--------| | README.md (## Architecture) | architecture | HIGH | docs/project/architecture.md | | docs/ARCHITECTURE.md | architecture | HIGH | docs/project/architecture.md | | CONTRIBUTING.md (## Development) | principles | MEDIUM | docs/principles.md | Cleanup Options: 1. NORMALIZE (recommended): Extract facts -> Archive -> Remove duplicate sources 2. ARCHIVE ONLY: Backup without extraction 3. SKIP: Leave non-canonical docs as-is (may cause duplication) Choose option (1/2/3): _
• If user selects "1" (NORMALIZE):

- Optional: "Review extracted facts before folding them into the canonical context? (yes/no)" - Confirm: "Proceed with cleanup and archive non-canonical files?"

• If user selects "2" (ARCHIVE ONLY):

- Confirm: "Archive non-canonical files to.archive/? Content will NOT be normalized."

• If user selects "3" (SKIP):

- Warn: "Non-canonical files will remain. This may cause duplication issues." - Proceed to Phase 1

0d. Backup and Archive:

• Create .archive/source-docs-{timestamp}/ directory
• Structure: `.archive/ - source-docs-YYYY-MM-DD-HHMMSS/ |- README_cleanup.md # Rollback instructions |- original/ # Exact copies of non-canonical files | |- README.md | |- ARCHITECTURE.md | - documentation/ - extracted/ # Extracted facts (for reference) |- architecture_content.md - principles_content.md `
• Copy all detected source files to original/
• Save extracted facts to extracted/
• Generate README_cleanup.md with rollback instructions

0e. Context Normalization:

• Build SOURCE_DOC_NOTES from extracted facts: {"SOURCE_DOC_NOTES": {"architecture_notes": {"sections": [...], "diagrams": [...]}, "requirements_notes": {"functional": [...]}, "principles_notes": {"principles": [...]}, "tech_stack_notes": {"frontend": "...", "backend": "..."}, "api_notes": {"endpoints": [...]}, "database_notes": {"tables": [...]}, "runbook_notes": {"install_steps": [...]}, "infrastructure_notes": {"servers": [...], "domains": [...], "ports": {}}}}
• Merge extracted facts into the coordinator input as supporting evidence only
• Workers consume normalized context fields and current source facts directly; there is no legacy-only branch
• Priority order: Current source facts + normalized project documentation > template defaults

0f. Cleanup (Non-Canonical Files):

• For root-level files (README.md, CONTRIBUTING.md):

- Do NOT delete - Remove duplicated sections using Edit tool - Add links to new locations: - ## Architecture -> replace with link: [Architecture](docs/project/architecture.md) - ## Tech Stack -> replace with link: [Tech Stack](docs/project/tech_stack.md)

• For standalone non-canonical files (ARCHITECTURE.md, documentation/):

- Delete files (already backed up) - Log: "Deleted: ARCHITECTURE.md (migrated to docs/project/architecture.md)"

• Clean empty non-canonical directories

Output: cleanup_summary {normalized_count, archived_count, skipped_count, source_doc_notes}

Phase 1: User Confirmation

Objective: Check existing files, explain workflow, and get user approval.

Process:

1. Cleanup Summary (if Phase 0 ran):

- Show: "Normalized {N} non-canonical documents" - Show: "Archived to.archive/source-docs-{date}/" - Show: "Coordinator input enriched with normalized source notes"

1. Pre-flight Check (scan existing documentation):

- Use Glob tool to check all potential files: - Root docs: AGENTS.md, CLAUDE.md, docs/README.md, docs/documentation_standards.md, docs/principles.md - Reference structure (5 items): docs/reference/README.md, docs/reference/adrs/, docs/reference/guides/, docs/reference/manuals/, docs/reference/research/ - Tasks docs (2 files): docs/tasks/README.md, docs/tasks/kanban_board.md - Project docs (up to 8 files): docs/project/requirements.md, architecture.md, tech_stack.md, api_spec.md, database_schema.md, design_guidelines.md, infrastructure.md, runbook.md - Test docs (2 files): docs/reference/guides/testing-strategy.md, tests/README.md - Count existing vs missing files - Show user summary: Documentation Status: Found: X existing files Missing: Y files Pipeline will create ONLY missing files. Existing files will be preserved (no overwrites).

1. Show user what will be created:

- Root + Project documentation (AGENTS.md canonical + CLAUDE.md import stub + docs/README.md + documentation_standards.md + principles.md + docs/project/ via ln-110-project-docs-coordinator) - Reference structure (docs/reference/ via ln-120-reference-docs-creator) - Task management docs (docs/tasks/ via ln-130-tasks-docs-creator) - Test documentation (tests/ via ln-140-test-docs-creator - optional) - Estimated time: 15-20 minutes with interactive dialog

1. Ask: "Proceed with creating missing files? (yes/no)"
2. Ask: "Include test documentation (tests/README.md)? (yes/no)"

Output: File scan summary + user approval + test docs preference

Phase 2: Invoke Coordinator + Workers Sequentially

Objective: Create complete documentation system by invoking L2 coordinator + 4 L2 workers in order.

Process (AUTOMATIC invocations with Skill tool):

2a. Create Root + Project Documentation:

• Invocation: Skill(skill: "ln-110-project-docs-coordinator") -> AUTOMATIC
• Input: Pass normalized source notes from Phase 0 when cleanup was performed
• Behavior: Coordinator gathers context ONCE, then delegates to 5 L3 workers:

- ln-111-root-docs-creator -> root docs (AGENTS.md canonical + CLAUDE.md stub + docs/README.md + documentation_standards.md + principles.md) - ln-112-project-core-creator -> 3 core docs - ln-113-backend-docs-creator -> 2 conditional - ln-114-frontend-docs-creator -> 1 conditional (if hasFrontend) - ln-115-devops-docs-creator -> 2 docs: 1 always + 1 conditional

• Output: Root docs (AGENTS.md canonical + CLAUDE.md @AGENTS.md import stub + docs/README.md + docs/documentation_standards.md + docs/principles.md) + Project docs (docs/project/requirements.md, architecture.md, tech_stack.md, infrastructure.md + conditional: api_spec.md, database_schema.md, design_guidelines.md, runbook.md)
• Store: Save context_store from ln-110 result (contains TECH_STACK for ln-120)
• Validation: Each L3 worker validates output (SCOPE, metadata markers, top sections, Maintenance)
• Verify: All documents exist before continuing

2b. Create Reference Structure + Smart Documents:

• Invocation: Skill(skill: "ln-120-reference-docs-creator") -> AUTOMATIC
• Input: Pass context_store from ln-110 (TECH_STACK enables smart document creation)
• Output: docs/reference/README.md + adrs/, guides/, manuals/, research/ directories + justified ADRs/Guides/Manuals
• Smart Creation: Creates documents only for nontrivial technology choices (see ln-120 justification rules)
• Validation: ln-120 validates output in Phase 2/3
• Verify: Reference hub exists before continuing

2c. Create Task Management Docs:

• Invocation: Skill(skill: "ln-130-tasks-docs-creator") -> AUTOMATIC
• Output: docs/tasks/README.md + optionally kanban_board.md (if user provides Linear config)
• Validation: ln-130 validates output in Phase 2/3
• Verify: Tasks README exists before continuing

2d. Create Test Documentation (Optional):

• Condition: If user approved test docs in Phase 1
• Invocation: Skill(skill: "ln-140-test-docs-creator") -> AUTOMATIC
• Output: tests/README.md (test documentation with Story-Level Test Task Pattern)
• Validation: ln-140 validates output in Phase 2/3
• Skip: If "no" -> can run ln-140-test-docs-creator later manually

2e. Extract Skills from Documentation (Optional):

• Condition: User approved skill extraction in Phase 1, or invoked manually later
• Invocation: Skill(skill: "ln-160-docs-skill-extractor") -> AUTOMATIC
• Input: All docs created by ln-110 through ln-140
• Output: .claude/commands/*.md files extracted from procedural documentation sections
• Skip: If not approved -> can run ln-160-docs-skill-extractor later manually

Output: Complete documentation system with coordinator + 4 workers completed and validated

Worker Invocation (MANDATORY)

TodoWrite format (mandatory)

Add ALL invocations to todos before starting:

- Invoke ln-110-project-docs-coordinator (pending)
- Invoke ln-120-reference-docs-creator (pending)
- Invoke ln-130-tasks-docs-creator (pending)
- Invoke ln-140-test-docs-creator (pending)
- Run Global Cleanup (Phase 3) (pending)
- Show Summary (Phase 4) (pending)

Mark each as in_progress when starting, completed when worker returns success.

Phase 3: Global Cleanup and Consolidation

Objective: Remove duplicates, orphaned files, consolidate knowledge across ALL documentation.

Trigger: Only after ALL workers complete Phase 2/3 validation.

Process:

3a. Build docs-quality manifest

Create one normalized manifest from worker outputs before any cleanup:

1. Collect created_files, skipped_files, and quality_inputs from ln-110, ln-120, ln-130, and ln-140
2. Merge quality_inputs.doc_paths and quality_inputs.owners, deduplicate paths, and fail fast if any worker omitted the normalized return contract
3. Write the explicit manifest with the shared verifier: node {skills_repo_root}/skills-catalog/shared/scripts/docs-quality/cli.mjs manifest \ --project-root. \ --files "AGENTS.md,CLAUDE.md,docs/README.md,docs/project/requirements.md" \ --output docs/project/.audit/ln-100/docs-quality-manifest.json
4. Keep the manifest as the Single Source of Truth for all Phase 3 quality decisions

3b. Run centralized docs-quality verifier

Run the shared verifier against the manifest:

node {skills_repo_root}/skills-catalog/shared/scripts/docs-quality/cli.mjs verify \
  --project-root . \
  --manifest docs/project/.audit/ln-100/docs-quality-manifest.json \
  --output docs/project/.audit/ln-100/docs-quality-report.json

Verifier checks come from the shared docs-quality contract/rules and include:

• SCOPE tags in required docs
• Maintenance section + markers
• Forbidden placeholders and leaked template metadata
• Broken internal links
• Disallowed code fences
• Stale inline dates
• Non-official stack documentation links

Gate: No HIGH/CRITICAL findings may remain before Phase 3 cleanup proceeds.

3c. Bounded repair loop

If verifier reports HIGH/CRITICAL findings:

1. Pass 1 - deterministic inline fixes:

- Add missing SCOPE tags - Add or complete ## Maintenance - Normalize broken internal links when target is obvious - Refresh stale generated dates - Remove leaked template metadata and unresolved markers - Replace illegal fenced code with tables, commands, or links

1. Pass 2 - owner-scoped semantic repair:

- Group remaining findings by quality_inputs.owners[path] - Re-invoke only the owning worker with the scoped findings for its files - Do not rerun unrelated workers

1. Re-run the verifier after each repair pass
2. Stop after 2 repair iterations total
3. If any HIGH/CRITICAL findings remain, fail the pipeline instead of emitting low-quality docs

Record the verifier summary and doc-registry output in the Phase 5 final output.

3d. Scan for duplicate content

1. Read all.md files in docs/

- Use Glob tool: pattern: "docs/**/*.md" - Store list of all documentation files

1. Identify duplicate sections:

- For each file: - Read file content - Extract section headers (##, ###) - Extract section content (text between headers) - Compare sections across files: - If 2+ sections have: - Same heading (case-insensitive) - 80% content similarity (simple word overlap check) - Mark as duplicate

1. Determine canonical version:

- Rules for canonical location: - "Development Principles" -> principles.md (always canonical) - "Testing Strategy" -> testing-strategy.md (always canonical) - "Linear Configuration" -> tasks/kanban_board.md (always canonical) - For other duplicates: Keep in FIRST file encountered (alphabetical order)

1. Remove duplicates:

- For each duplicate section: - Use Edit tool to delete from non-canonical files - Use Edit tool to add link to canonical location: Canonical source: [Development Principles](../principles.md#development-principles) - Track count of removed duplicates

1. Log results:

- "Removed {count} duplicate sections" - List: "{section_name} removed from {file} (canonical: {canonical_file})"

3e. Report unexpected files (advisory)

1. List all.md files in docs/

- Use Glob tool: pattern: "docs/**/*.md"

1. Check against expected structure (files created by workers + user-created reference docs)
2. Report findings (DO NOT move/delete/archive):

- List unexpected files with advisory message - User decides what to do with them - Log: "{count} unexpected files found (not in expected structure) -> listed for user review"

3f. Consolidate knowledge

1. Identify scattered information:

- Known patterns: - Linear config scattered: kanban_board.md + tasks/README.md - Development principles scattered: AGENTS.md + principles.md + architecture.md - Testing info scattered: testing-strategy.md + tests/README.md + architecture.md

1. For each scattered concept:

- Determine Single Source of Truth (SSoT): - Linear config -> kanban_board.md - Development principles -> principles.md - Testing strategy -> testing-strategy.md

1. Update non-SSoT files:

- Use Edit tool to replace duplicate content with link: Canonical source: [Linear Configuration](../tasks/kanban_board.md#linear-configuration) - Track consolidation count

1. Log results:

- "Consolidated {count} scattered concepts" - List: "{concept} consolidated to {SSoT_file}"

3g. Cross-link validation

1. Scan all.md files for internal links:

- For each file: - Read content - Extract all markdown links: [text](path) - Filter internal links (relative paths, not http://)

1. Verify link targets exist:

- For each link: - Resolve relative path - Check if target file exists (Glob tool) - If missing: Mark as broken

1. Fix broken links:

- For each broken link: - If target was archived: Update link to archive path - If target never existed: Remove link (convert to plain text) - Track fix count

1. Add missing critical links:

- AGENTS.md -> docs/README.md: - Read AGENTS.md - Check for link to docs/README.md - If missing: Add in "Documentation Hub" section - docs/README.md -> section READMEs: - Check for links to project/, reference/, tasks/, tests/ READMEs - If missing: Add - Track added links count

1. Log results:

- "Fixed {count} broken links" - "Added {count} missing critical links" - List changes

3h. Final report

Global Cleanup Complete:

Structure:
- Removed {N} duplicate sections (canonical: principles.md)
- Found {N} unexpected files (listed for user review)
- Consolidated {N} scattered concepts

Links:
- Fixed {N} broken links
- Added {N} missing critical links:
  - list of added links

Output: All documentation cleaned up, duplicates removed, unexpected files reported, knowledge consolidated, cross-links validated

Phase 4: Summary and Next Steps

Objective: Provide complete overview of created system.

Process:

1. List all created files with sizes:

- AGENTS.md (canonical project entry point) - CLAUDE.md (@AGENTS.md import stub for Claude Code) - docs/README.md (root documentation hub) - docs/documentation_standards.md (60 universal requirements) - docs/principles.md (11 development principles) - docs/project/requirements.md, architecture.md, tech_stack.md + conditional documents (3-7 total) - docs/reference/README.md (reference hub with empty adrs/, guides/, manuals/, research/ directories) - docs/tasks/README.md + optionally kanban_board.md - tests/README.md (if created)

1. Show documentation system features:

- SCOPE tags (document boundaries defined) - Maintenance sections (update triggers + verification) - README hub (central navigation) - DAG structure (Directed Acyclic Graph navigation) - Living documentation ready - Deduplicated content (canonical sources only) - Validated cross-links (no broken links)

1. Recommend next steps:

- "Review generated documentation (AGENTS.md canonical + CLAUDE.md stub -> docs/)" - "Run ln-210-epic-coordinator to decompose scope into Epics" - "Share documentation with technical stakeholders"

1. Summarize the docs-quality gate:

- Verifier score: X/10 - Repair passes used: 0-2 - Remaining warnings: list LOW/MEDIUM only

Output: Summary message with file list, verifier status, and recommendations

Complete Output Structure

project_root/
|- AGENTS.md                         # Canonical project entry point (map-first root)
|- CLAUDE.md                         # Derived Anthropic entrypoint
|- docs/
|  |- README.md                     # Root documentation hub (general standards)
|  |- documentation_standards.md    # 60 universal requirements (Claude Code + industry standards)
|  |- principles.md                 # 11 development principles (Standards First, YAGNI, KISS, DRY, etc.)
|  |- project/
|  |  |- requirements.md           # Functional Requirements (NO NFR per project policy)
|  |  |- architecture.md           # arc42-based architecture with C4 Model
|  |  |- tech_stack.md             # Technology versions, Docker config
|  |  |- api_spec.md               # API endpoints (conditional)
|  |  |- database_schema.md        # Database schema (conditional)
|  |  |- design_guidelines.md      # UI/UX system (conditional)
|  |  |- infrastructure.md         # Infrastructure inventory (always)
|  |  `- runbook.md                # Operations guide (conditional)
|  |- reference/
|  |  |- README.md                 # Reference documentation hub (registries)
|  |  |- adrs/                     # Empty, ready for ADRs
|  |  |- guides/                   # Empty, ready for guides
|  |  |- manuals/                  # Empty, ready for manuals
|  |  `- research/                 # Empty, ready for research
|  |- tasks/
|  |  |- README.md                 # Task management system rules
|  |  `- kanban_board.md           # Linear integration (optional)
`- tests/
   `- README.md                    # Test documentation (optional)

Integration with Project Workflow

Recommended workflow for new projects:

1. ln-100-documents-pipeline (this skill) - Create complete documentation system
2. ln-210-epic-coordinator - Decompose scope into Epics (Linear Projects)
3. ln-220-story-coordinator - Create User Stories for each Epic (automatic decomposition + replan)
4. ln-300-task-coordinator - Break down Stories into implementation tasks (automatic decomposition + replan)
5. ln-310-multi-agent-validator - Verify Stories before development
6. ln-400-story-executor - Orchestrate Story implementation
7. Story quality gate - Review completed Stories

Trade-offs

Trade-offs:

• Less granular control (can't skip coordinator phases)
• Longer execution time (15-20 minutes)
• Global cleanup runs even if only one file was created

When to use manual approach instead:

• Need one specific ADR/guide/manual -> use shared/templates/ + shared/references/documentation_creation.md

Documentation Standards

All documents created by this pipeline MUST follow these rules:

These standards are enforced by L3 workers (ln-111-115), ln-120/130/140, and the shared docs-quality gate.

Error Handling

If any invoked skill fails:

1. Notify user which skill failed
2. Show error message from failed skill
3. Recommend manual invocation for debugging
4. List already completed steps (partial progress)

Technical Implementation Notes

Skill Invocation:

• Uses Skill tool with command parameter
• Waits for each skill to complete before proceeding
• Verifies output files exist before moving to next phase

File Verification:

• Uses Glob tool to check docs/project/ structure
• Lists file sizes for user confirmation
• Warns if expected files missing

Global Cleanup:

• Uses Glob tool to find all.md files
• Uses Read tool to analyze content
• Uses Edit tool to remove duplicates and add links
• Reports unexpected files (advisory, no auto-archive)

Docs-Quality Gate:

• Uses shared/scripts/docs-quality/cli.mjs to build a manifest from worker outputs
• Uses the shared verifier before cleanup and again after repair passes
• Fails fast on unresolved HIGH/CRITICAL findings instead of emitting low-quality documentation

Standards Compliance:

• All output follows same standards as underlying skills
• ISO/IEC/IEEE 29148:2018 (Requirements)
• ISO/IEC/IEEE 42010:2022 (Architecture)
• arc42 + C4 Model + Michael Nygard's ADR Format

Critical Rules

• Idempotent: Creates only missing files; existing files are preserved without overwrite
• Sequential invocation: Workers must be invoked in order (ln-110 -> ln-120 -> ln-130 -> ln-140); each verified before next
• Docs-quality gate mandatory: Phase 3 must build manifest -> run verifier -> execute bounded repair loop before cleanup continues
• Global cleanup mandatory: Phase 3 (deduplication, orphan reporting, SSoT consolidation, cross-link validation) runs only after docs-quality gate passes
• User confirmation required: Pre-flight check and explicit approval before any file creation
• NO_CODE Rule: All generated documents use tables/ASCII/links; no code blocks >5 lines

Phase 5: Meta-Analysis

MANDATORY READ: Load shared/references/meta_analysis_protocol.md

Skill type: planning-coordinator. Run after all phases complete. Output to chat using the planning-coordinator format.

Reference Files

• Non-canonical source patterns: references/legacy_detection_patterns.md
• Worker skills: ln-110-project-docs-coordinator, ln-120-reference-docs-creator, ln-130-tasks-docs-creator, ln-140-test-docs-creator
• Shared docs-quality contract: shared/references/docs_quality_contract.md
• Shared verifier CLI: shared/scripts/docs-quality/cli.mjs

Definition of Done

Before completing work, verify ALL checkpoints:

Source Cleanup (Phase 0 - if applicable):

• Non-canonical documentation patterns applied (Glob + Grep)
• Source manifest built: {path, type, confidence, target}
• User selected cleanup option (NORMALIZE / ARCHIVE / SKIP)
• If NORMALIZE: facts extracted using type-specific extractors
• Backup created: .archive/source-docs-{timestamp}/original/
• Extracted facts saved: .archive/source-docs-{timestamp}/extracted/
• README_cleanup.md generated with rollback instructions
• SOURCE_DOC_NOTES prepared for coordinator input
• Non-canonical files cleaned up (sections removed from README.md, standalone files deleted)

User Confirmation (Phase 1):

• Migration summary shown (if Phase 0 ran)
• Workflow explained to user (coordinator + 4 workers: ln-110 -> ln-120 -> ln-130 -> ln-140)
• User approved: "Proceed with complete documentation system creation? (yes/no)"
• Test docs preference captured: "Include test documentation? (yes/no)"

Coordinator + Workers Invoked Sequentially (Phase 2):

• ln-110-project-docs-coordinator invoked -> Output verified: Root docs (AGENTS.md + CLAUDE.md + docs/README.md + docs/documentation_standards.md + docs/principles.md) + Project docs (docs/project/requirements.md, architecture.md, tech_stack.md + conditional files)
• ln-120-reference-docs-creator invoked -> Output verified: docs/reference/README.md + directories (adrs/, guides/, manuals/, research/) + justified ADRs/Guides/Manuals based on TECH_STACK
• ln-130-tasks-docs-creator invoked -> Output verified: docs/tasks/README.md + optionally kanban_board.md
• ln-140-test-docs-creator invoked (if enabled) -> Output verified: tests/README.md
• Each component validated its own output (SCOPE, metadata markers, top sections, Maintenance, POSIX compliance)
• Each worker returned normalized created_files, skipped_files, quality_inputs, and validation_status

File Verification Complete:

• All expected files exist (Glob tool used to verify structure)
• File sizes listed for user confirmation
• Warning displayed if expected files missing

Global Cleanup Complete (Phase 3):

• 3a: Manifest built from normalized worker outputs
• 3b: Shared verifier executed against the manifest
• 3c: Bounded repair loop completed; no HIGH/CRITICAL findings remain
• 3d: Duplicate sections identified and removed (>80% similarity)
• 3d: Links added to canonical locations (principles.md, testing-strategy.md, kanban_board.md)
• 3e: Unexpected files reported (advisory, no auto-archive)
• 3f: Scattered concepts consolidated to Single Source of Truth (SSoT)
• 3g: Internal links validated (broken links fixed, critical links added)
• 3h: Final report generated (quality gate summary, counts, lists, actions)

Summary Displayed (Phase 4):

• All created files listed with sizes
• Documentation system features highlighted (SCOPE, metadata markers, Quick Navigation, Agent Entry, Maintenance, README hubs, DAG structure, deduplicated content, validated links)
• Docs-quality verifier summary shown (score, repair passes, remaining LOW/MEDIUM warnings)
• Next steps recommended (ln-210-epic-coordinator)

Error Handling (if applicable):

• If any worker failed: User notified which worker failed, error message shown, manual invocation recommended, partial progress listed

Output: Complete documentation system (AGENTS.md canonical + CLAUDE.md @AGENTS.md import stub + docs/ with README.md, documentation_standards.md, principles.md + optionally tests/) with global cleanup (no duplicates, orphaned files reported, consolidated knowledge, validated cross-links)

Version: 8.1.0 Last Updated: 2025-01-12