ln-161-skill-creator（ln 161 技能创建器）

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}.

ln-161-skill-creator

Type: L3 Worker (standalone-capable) Category: 1XX Documentation Pipeline

Creates .claude/commands/*.md slash command files from procedural documentation sections. It converts procedural prose into imperative command instructions without changing the source docs.

Overview

Input Modes

Standalone Discovery

When invoked without approved_sections:

1. Read shared/references/procedural_extraction_rules.md, shared/references/procedural_skill_sop_guide.md, shared/references/markdown_read_protocol.md, and shared/references/docs_quality_contract.md
2. Scan docs-first sources:

- docs/**/*.md - tests/README.md - tests/manual/**/* - README.md - CONTRIBUTING.md

1. Use AGENTS.md, CLAUDE.md, and docs/project/.context/doc_registry.json only to route discovery
2. Classify candidate sections with the shared procedural extraction rules
3. Present the extraction plan to the user for approval
4. Create approved commands

Workflow

Phase 1: Prepare

Receive or build this normalized input:

{
  "approved_sections": [
    {
      "source_file": "docs/project/runbook.md",
      "section_header": "Deployment",
      "line_range": [45, 92],
      "command_name": "deploy.md",
      "doc_kind": "how-to",
      "doc_role": "canonical"
    }
  ]
}

MANDATORY READ: Load references/command_template.md and shared/references/procedural_skill_sop_guide.md

Phase 2: Transform and Create

For each approved section:

1. Read the source section using the shared markdown read protocol
2. Ignore standardized doc shell content if it appears in the selected range:

- header markers such as DOC_KIND, DOC_ROLE, READ_WHEN, SKIP_WHEN, PRIMARY_SOURCES - Quick Navigation - Agent Entry - Maintenance

1. Detect allowed-tools

1. Transform the content with these rules:

1. Write .claude/commands/{command_name} if the file does not already exist

Phase 3: Report

Return:

created:
  - file: .claude/commands/deploy.md
    source: docs/project/runbook.md#Deployment
    lines: 85
    tools: [Bash, Read]
summary: "Created 1 command from 1 procedural section"

Critical Rules

• Template-driven: All output follows references/command_template.md.
• Preserve source: Never modify or delete source docs.
• No invention: Do not add commands, steps, or paths absent from the source.
• Imperative voice: Every retained instruction must be actionable.
• SOP/TWI extraction: Preserve action, decision point, exception, validation, dependency/preflight, and why for risky procedural steps.
• No copied doc shell: Do not copy DOC_KIND, DOC_ROLE, Quick Navigation, Agent Entry, or Maintenance into commands.
• Source provenance: Every generated command must point back to its source doc and section.
• Relative paths: File references must stay relative to project root.
• Idempotent: Skip existing command files; do not overwrite them.

Definition of Done

• Approved sections received or discovered
• Source sections read with section-first protocol
• Allowed-tools detected per command
• Content transformed into imperative workflow steps
• Risky procedural steps include action, key point, why, evidence, exception, and automation/guard where applicable
• Standard doc shell content removed from command output
• Source provenance included in each created command
• Existing command files not overwritten
• Creation report returned

Version: 1.0.0 Last Updated: 2026-03-26