Skill Factory

Overview

Meta-skill that generates complete, standards-compliant Claude skills through an interactive workflow. Produces production-ready packages with SKILL.md, references, scripts, and automated triggering tests.

Quick Decision Tree

┌─────────────────────────────────────────────────────────────┐
│                    What do you need?                        │
└─────────────────────────────────────────────────────────────┘
                            │
        ┌───────────────────┼───────────────────┐
        ▼                   ▼                   ▼
┌───────────────┐   ┌───────────────┐   ┌───────────────┐
│ NEW SKILL     │   │ MIGRATION     │   │ IMPROVEMENT   │
│ from scratch  │   │ prompt → skill│   │ existing skill│
└───────┬───────┘   └───────┬───────┘   └───────┬───────┘
        │                   │                   │
        ▼                   ▼                   ▼
   Phase 1-6            Phase 2-6            Audit + 
   (full flow)       (skip analysis)      suggestions

Interactive Workflow (6 Phases)

Phase 1: Pre-Creation Analysis

Goal: Validate skill necessity and define scope

Questions to answer:

1. What problem does this skill solve? (1 sentence)
2. How often is this task performed? (≥5 past, ≥10 future = good candidate)
3. Who is the target persona? (dev backend, data analyst, PM...)
4. What trigger words will users employ?
5. What are the success criteria? (measurable output)
6. What is explicitly OUT of scope?

Output: Completed analysis using pre-analysis.md

Decision gate:

• ✅ Proceed if: recurring task, stable procedures, clear scope
• ❌ Stop if: one-time task, volatile procedures, duplicates existing capability

---

Phase 2: Skill Architecture Design

Goal: Define structure and file organization

Decisions:

1. Skill name: kebab-case, ≤64 chars, descriptive
2. Target platforms: Claude.ai / Claude Code / Both
3. Complexity level:
   • Simple: SKILL.md only
   • Standard: SKILL.md + references/
   • Advanced: Full package with scripts + templates
4. Multi-workflow?: If yes → design decision tree

Output: Structure diagram + file list

Reference: skill-structure.md

---

Phase 3: Description Engineering (Triggering)

Goal: Craft optimal description for semantic matching

Formula:

[CAPABILITIES] + [USE CASES] + [TRIGGERS] + [BOUNDARIES]

Checklist:

• Action verbs (extract, analyze, create, validate...)
• File types / data types concerned
• "Use when..." with 2-3 contexts
• "Not for..." with explicit exclusions
• No overlap with existing skills

Reference: description-formulas.md

Validation: Description must be ≤1024 chars

---

Phase 4: Workflow & Instructions Design

Goal: Define the operational workflow

Structure:

1. Overview (2-3 sentences)
2. Quick Start / Decision Tree (if multi-workflow)
3. Numbered workflow steps
4. Critical rules (mandatory filters, validations)
5. Examples (input → output)
6. Links to references
7. Explicit limitations

Guidelines:

• Instructions < 5000 tokens
• Progressive disclosure: details in references/
• Every referenced file must have explicit link

---

Phase 5: Validation (Dry-Run)

Goal: Verify before generation

Preview includes:

• Final structure diagram
• Complete SKILL.md preview
• Generated test cases
• Checklist compliance report

Checklist verification (12 points): → See checklist-validation.md

User approval required before Phase 6

---

Phase 6: Generation

Goal: Produce complete skill package

Outputs generated:

1. skill-name/SKILL.md - Main skill file
2. skill-name/references/*.md - Reference documentation
3. skill-name/scripts/test_triggering.py - Automated tests
4. skill-name/templates/* - Output templates (if needed)

Post-generation:

• Compliance report
• Deployment instructions
• Suggested test queries

---

Critical Cheatsheet (Inline)

YAML Frontmatter Rules

---
name: kebab-case-max-64-chars      # REQUIRED
description: >-                    # REQUIRED, max 1024 chars
  Action verbs + file types + "Use when..." + "Not for..."
allowed-tools: Read, Grep, Glob    # OPTIONAL (Claude Code only)
---

Description Formula

[WHAT it does] + [WHAT files/data] + [WHEN to use] + [WHEN NOT to use]

Structure Limits

Element	Limit
name	≤64 chars, kebab-case
description	≤1024 chars
SKILL.md body	<5000 tokens
Folder depth	Max 2 levels

Token Budget

Level	Content	Tokens
L1	Frontmatter	~100 (always loaded)
L2	SKILL.md body	<5000 (on trigger)
L3	References	Variable (on demand)

Anti-Patterns to Avoid

• ❌ Vague descriptions → random triggering
• ❌ Everything in SKILL.md → context overflow
• ❌ Unlinked reference files → never loaded
• ❌ Multi-purpose skills → split into focused skills
• ❌ Tabs in YAML → use spaces only

---

Knowledge Base

References

• Best Practices Guide - Complete methodology
• Description Formulas - Triggering patterns
• YAML Rules - Syntax validation
• Validation Checklist - 12-point verification
• Domain Templates - Pre-configured templates

Templates

• Pre-Analysis Template
• Skill Structure Template

Scripts

• Triggering Test Template

---

Supported Domains (Templates Available)

Domain	Typical Use Cases
SQL/Data	Data warehouse queries, KPI analysis, reporting
Code Review	Security audit, best practices, refactoring
Documentation	Technical docs, API reference, guides
File Processing	PDF, Excel, CSV manipulation
Workflow Automation	Multi-step business processes

---

Limitations

This skill does NOT:

• Generate skills for highly volatile procedures
• Create skills duplicating Claude's native capabilities
• Build skills requiring external API integrations (use MCP instead)
• Generate skills in languages other than English (skill content only)

---

Version History

Version	Date	Changes
1.0.0	2025-12-10	Initial release

Current: v1.0.0