Skill Creator

Build new Claude skills from scratch or update existing ones to meet Skillmark benchmark standards with proper spec files and test coverage.

What This Skill Does

The Challenge: Creating high-quality Claude skills requires understanding the SKILL.md specification format, prompt template conventions, activation patterns, and Skillmark benchmark requirements. Poorly structured skills fail benchmarks and confuse users.

The Solution: Skill Creator automates scaffolding of new skills with proper SKILL.md format, example prompts, configuration files, and benchmark-compatible test cases. Works for creating new skills or auditing/updating existing ones.

Activation

Implicit: Activates when creating new Claude skills or modifying existing skill definitions.

Explicit: Activate via prompt:

Activate skill-creator skill to build a new skill for automating social media scheduling

Capabilities

1. SKILL.md Scaffolding

Generate properly structured skill definition files.

SKILL.md structure:

# Skill Name

## Overview
Brief description of what this skill does and when it activates.

## Activation
- Implicit: Conditions for automatic activation
- Explicit: Activation command and syntax

## Capabilities
List of specific capabilities with examples.

## Configuration
Required env vars, files, dependencies.

## Examples
Input/output examples for each capability.

## Benchmarks
Skillmark test cases and expected outputs.

2. Prompt Template Creation

Design effective system prompts and user-facing activation syntax.

Prompt design principles:

  • Clear activation trigger (explicit vs implicit)
  • Defined output format
  • Error handling instructions
  • Scope boundaries (what skill does NOT do)

3. Test Case Generation

Create Skillmark-compatible benchmark test cases.

Test case format:

{
  "skill": "skill-name",
  "input": "User prompt that should activate skill",
  "expected_behavior": "Description of correct response",
  "evaluation_criteria": [
    "Uses correct output format",
    "Handles error case X",
    "Produces result within scope"
  ]
}

4. Skill Audit

Review existing skills against current SKILL.md spec and benchmark standards.

Audit checklist:

  • SKILL.md format compliance
  • Activation patterns documented
  • Examples cover all capabilities
  • Error handling specified
  • Test cases present

Prerequisites

  • Target skill purpose and capabilities defined
  • Access to Skillmark benchmark requirements
  • Existing skills directory (for updates)

Configuration

Skills directory location:

~/.claude/skills/
├── skill-name/
│   ├── SKILL.md          # Main skill definition
│   ├── prompts/          # Prompt templates
│   ├── examples/         # Input/output examples
│   └── tests/            # Benchmark test cases

Best Practices

1. One skill, one responsibility Skills should do one thing well. If a skill needs more than 5 capabilities, split it.

2. Write examples before prompts Define what good output looks like, then write prompts that produce it.

3. Test edge cases explicitly Skillmark benchmarks focus on edge cases. Document how the skill handles malformed input, missing context, and ambiguous requests.

Common Use Cases

Use Case 1: New Tool Integration Skill

Scenario: Create a skill for a new API tool (e.g., “search-web” using Perplexity API).

Workflow:

  1. Define skill purpose and activation triggers
  2. Scaffold SKILL.md with capabilities
  3. Write example prompts and expected outputs
  4. Create configuration template
  5. Generate benchmark test cases

Output: Complete skill package ready for ~/.claude/skills/.

Use Case 2: Skill Audit and Update

Scenario: Existing skill fails Skillmark benchmarks after spec update.

Workflow:

  1. Read current SKILL.md
  2. Compare against latest spec requirements
  3. Identify gaps (missing examples, outdated format)
  4. Update SKILL.md to current format
  5. Add missing test cases

Output: Updated skill package with benchmark compliance.

Troubleshooting

Issue: Skill activates unexpectedly in wrong contexts Solution: Narrow implicit activation conditions. Add explicit “does not activate when…” clause.

Issue: Benchmark test cases fail despite correct output Solution: Review evaluation criteria — may need more specific output format requirements.

  • /ckm:skill-creator - Create or update skills
  • /ckm:template-skill - Minimal skill scaffold