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.

ln-160-docs-skill-extractor

Type: L2 Coordinator Category: 1XX Documentation Pipeline Workers: ln-161-skill-creator, ln-162-skill-reviewer

Scans project documentation, identifies procedural content (deploy, test, SSH, troubleshoot), and extracts it into executable .claude/commands/*.md skills. Declarative content (architecture, requirements, API specs) remains as documentation.

---

Overview

Aspect	Details
Input	Project docs (docs/, tests/, README.md)
Output	.claude/commands/*.md files in target project
Workers	ln-161 (create commands), ln-162 (review commands)

---

Workflow

Phase 1: Discovery (scan docs, build inventory)
    |
Phase 2: Classification (procedural vs declarative scoring)
    |
Phase 3: Extraction Plan (user approval)
    |
Phase 4: Delegate -> ln-161 (create commands)
    |
Phase 5: Delegate -> ln-162 (review commands)
    |
Phase 6: Report (aggregate results)

---

Phase 1: Discovery

Scan target project documentation to build content inventory.

Scan targets:

• docs/ (all .md recursively)
• tests/README.md, tests/manual/
• .claude/commands/ (existing -- to avoid duplicates)
• README.md, CONTRIBUTING.md

Per file: Extract H2/H3 sections with metadata.

Build contextStore:

contextStore:
  project_root: {CWD}
  existing_commands: [list of .claude/commands/*.md filenames]
  doc_inventory:
    - file: docs/project/runbook.md
      sections:
        - header: "Deployment"
          line_range: [45, 92]
          signals: { code_blocks: 3, numbered_steps: 5, imperative_verbs: 8 }

---

Phase 2: Classification

MANDATORY READ: Load references/classification_rules.md

Score each section as PROCEDURAL vs DECLARATIVE via weighted signals. Apply thresholds to classify:

Classification	Condition	Action
PROCEDURAL	proc >= 4 AND proc > decl * 2	Extract to command
DECLARATIVE	decl >= 4 AND decl > proc * 2	Skip (keep as doc)
MIXED	Both >= 3	Partial extraction
THIN	Both < 3	Skip

Filter: remove sections already covered by existing .claude/commands/.

---

Phase 3: Extraction Plan (User Approval Gate)

Present classified results via AskUserQuestion:

Found {N} procedural sections in {M} files:

| # | Source | Section | Score | Proposed Command |
|---|--------|---------|-------|------------------|
| 1 | runbook.md | Deployment | P:8/D:1 | deploy.md |
| 2 | runbook.md | Troubleshooting | P:6/D:0 | troubleshoot.md |
| 3 | tests/README.md | Running Tests | P:7/D:2 | run-tests.md |

Existing .claude/commands/ (will skip): refresh_context.md, build-and-test.md

Include? (e.g., "1,2,3" or "all" or "all skip 3")

If user approves none, end with "No skills to create."

---

Phase 4: Delegate to ln-161 (Skill Creation)

Pass approved sections from contextStore to ln-161-skill-creator:

Agent(
  description: "Create commands from docs",
  prompt: "Execute skill creator.\nStep 1: Invoke:\n  Skill(skill: \"ln-161-skill-creator\")\nCONTEXT:\n{contextStore with approved sections}",
  subagent_type: "general-purpose"
)

Collect: list of created file paths + per-file summary.

---

Phase 5: Delegate to ln-162 (Skill Review)

Pass created command file paths to ln-162-skill-reviewer:

Agent(
  description: "Review created commands",
  prompt: "Execute skill reviewer in COMMAND mode.\nStep 1: Invoke:\n  Skill(skill: \"ln-162-skill-reviewer\", args: \"commands\")\nFILES: {list of created paths}",
  subagent_type: "general-purpose"
)

Collect: review verdicts per file + aggregate pass rate.

---

Phase 6: Report

Aggregate results from ln-161 (created files) + ln-162 (review verdicts):

## Docs Skill Extractor -- Complete

| Metric | Count |
|--------|-------|
| Documents scanned | {N} |
| Sections analyzed | {N} |
| Procedural found | {N} |
| Commands created | {N} |
| Commands skipped (existing) | {N} |
| Review PASS | {N} |
| Review FIXED | {N} |
| Review WARN | {N} |

Created commands:
- .claude/commands/deploy.md (from runbook.md#Deployment)
- .claude/commands/run-tests.md (from tests/README.md#Running Tests)

Next steps:
- Test each command by invoking /{command-name}
- Customize generated commands for project-specific needs

---

Critical Rules

• Detect before extract: Only extract sections scoring PROCEDURAL -- never force extraction on declarative content
• No duplicates: Skip sections already covered by existing .claude/commands/ files
• User approval required: Never create commands without Phase 3 confirmation
• Delegate creation: All command file writing goes through ln-161 (SRP)
• Delegate review: All command validation goes through ln-162 (SRP)
• Preserve docs: Source documentation stays intact -- commands are extracted, not moved

---

Definition of Done

• Doc inventory built (all scan targets discovered)
• Classification scored for every section (PROCEDURAL/DECLARATIVE/MIXED/THIN)
• User approved extraction plan (Phase 3 gate passed)
• ln-161 created all approved commands
• ln-162 reviewed all created commands
• Report aggregated with scan/create/review metrics

---

Phase 7: Meta-Analysis

MANDATORY READ: Load shared/references/meta_analysis_protocol.md

Skill type: planning-coordinator. Run after Phase 6 completes. Output to chat using the planning-coordinator format.

---

Version: 1.0.0 Last Updated: 2026-03-13