Platxa Skill Generator

Platxa Skill Generator

Create Claude Code skills autonomously using a multi-phase orchestrated workflow.

Overview

This skill guides you through creating production-ready Claude Code skills by:

1. Discovering domain knowledge through web research and existing skill analysis
2. Architecting the skill structure based on type (Builder/Guide/Automation/Analyzer/Validator)
3. Generating SKILL.md content, scripts, and reference documentation
4. Validating against Anthropic's Agent Skills spec with quality scoring

The workflow uses Task tool subagents for each phase, ensuring deep expertise at every step.

Workflow

Phase 1: Initialize

User: /skill-generator

Ask the user for:

• Skill description (what should the skill do?)
• Target users (who will use this skill?)

Phase 2: Discovery (Automatic)

Use Task tool with subagent_type="Explore" to:

1. Search web for domain best practices
2. Fetch relevant documentation
3. Analyze existing skills in ~/.claude/skills/ for patterns
4. Identify procedural knowledge (HOW) vs domain expertise (WHAT)
5. Determine what varies by project vs what's constant

Sufficiency Check: Evaluate if research is complete. Only ask user for clarification if gaps exist.

Phase 3: Architecture

Based on discovery, determine:

• Skill Type: Builder, Guide, Automation, Analyzer, or Validator
• Structure: Which directories needed (scripts/, references/, assets/)
• Token Budget: SKILL.md < 500 lines, metadata ~100 tokens

Generate architecture blueprint JSON:

{
  "skill_type": "Builder",
  "directories": ["scripts", "references"],
  "skill_md_sections": ["Overview", "Workflow", "Examples", "Output Checklist"],
  "scripts": [{"name": "validate.sh", "purpose": "Validate output"}],
  "references": [{"name": "best-practices.md", "purpose": "Domain expertise"}]
}

Phase 4: Generation

Create all skill files:

1. SKILL.md with valid YAML frontmatter
2. scripts/ with executable helpers
3. references/ with domain documentation

Use templates from references/templates/ based on skill type.

Phase 5: Validation

Run quality checks:

• SKILL.md exists with valid frontmatter
• Name is hyphen-case, ≤64 characters
• Description is ≤1024 characters
• Token budget within limits
• Scripts are executable and secure
• References provide real domain expertise
• Quality score ≥7.0/10 (python3 scripts/score-skill.py <dir> --json)

Quality Score: Run score-skill.py for 5-dimension assessment (spec compliance, content depth, example quality, structure, token efficiency). Must be ≥7.0/10 to pass.

Evaluation Scaffold: After passing validation, generate 3 eval scenarios (happy-path, edge-case, error-handling) in evals/evals.json for testing the skill's real-world behavior.

Phase 6: Installation

Ask user for installation location:

• User skill: ~/.claude/skills/{skill-name}/
• Project skill: .claude/skills/{skill-name}/

Copy files and verify installation.

Examples

Example 1: Creating a Code Documenter Skill

User: /skill-generator
Assistant: What skill would you like to create?
User: A skill that generates documentation for Python code
Assistant: [Runs Discovery subagent to research Python docstring standards...]
Assistant: [Creates architecture for "Automation" type skill...]
Assistant: [Generates py-documenter skill with SKILL.md, scripts/generate-docs.sh]
Assistant: Quality score: 8.2/10. Install to ~/.claude/skills/py-documenter? (y/n)

Example 2: Creating a Security Analyzer Skill

User: /skill-generator
Assistant: What skill would you like to create?
User: A skill that scans code for security vulnerabilities
Assistant: [Discovers OWASP Top 10, SAST patterns, CVE databases...]
Assistant: [Creates "Analyzer" type skill with references/owasp-checklist.md]
Assistant: Quality score: 7.8/10. Ready to install.

Output Checklist

When generating a skill, verify:

• SKILL.md has valid YAML frontmatter
• Name follows hyphen-case (e.g., my-skill-name)
• Name is ≤64 characters
• Description is ≤1024 characters
• allowed-tools lists only tools the skill needs
• Workflow section has clear steps
• Examples show realistic usage
• References contain actual domain expertise (not placeholders)
• Scripts are executable and tested
• Quality score ≥7.0/10

Skill Types Reference