Skill Maker 🔨

Create powerful, reusable skills with structured reasoning.

The Skill Maker Framework

CODEBLOCK0

Decision Tree: What Are We Building?

CODEBLOCK1

Step 1: Interpret

The Core Questions

Self-Check: Interpretation

• - [ ] Can I describe the skill in one sentence?
• [ ] Do I know what phrases would trigger it?
• [ ] Is this truly a new capability?

Step 2: Design

Complexity Decision

CODEBLOCK2

Directory Structure

CODEBLOCK3

Writing Triggers

Users typically say:

• - "I need to [action]"
• "How do I [task]?"
• "Help me with [domain]"
• "Can you [capability]?"

Formula for description:

"[What it does]. Use when: (1) [situation 1], (2) [situation 2], (3) [situation 3]."

Example:

"Fetch weather data from wttr.in. Use when: (1) User asks about weather, (2) User wants forecast, (3) User asks temperature in [city]."

Self-Check: Design

• - [ ] Name follows convention (lowercase, hyphens)?
• [ ] Description has clear triggers?
• [ ] I know which resources to include?

Step 3: Forge

SKILL.md Template

Copy this template for your skill:

CODEBLOCK4

Content Patterns

Progressive Disclosure

CODEBLOCK5

Self-Check: Forge

• - [ ] Frontmatter complete (name + description)?
• [ ] Body has reasoning framework?
• [ ] Self-check prompts included?
• [ ] Resources properly structured?

Step 4: Test

Trigger Testing

Read your description and ask:

CODEBLOCK6

Self-Check: Test

• - [ ] Does description match likely user phrases?
• [ ] Is the skill findable via search?
• [ ] Are there clear steps to follow?
• [ ] Does it include error handling?

Step 5: Polish

Refinement Loop

CODEBLOCK7

Common Fixes

Self-Check: Polish

• - [ ] Tested on real task?
• [ ] User feedback incorporated?
• [ ] Ready for regular use?

Versioning Guide

When to Bump Version

Changelog Format

CODEBLOCK8

Self-Check: Versioning

• - [ ] Did I increment the version correctly?
• [ ] Is changelog updated?
• [ ] Is this a breaking change?

Metadata Best Practices

Frontmatter Fields

CODEBLOCK9

Emoji Selection

Category Tags

Requirements Metadata

CODEBLOCK10

Self-Check: Metadata

• - [ ] Is frontmatter complete?
• [ ] Is emoji appropriate for category?
• [ ] Are requirements listed?
• [ ] Is version correct?

Why This Works

The Skill Logic Pattern

Based on research (SkillsBench 2026):

1. 1. Reasoning framework → Agent knows HOW to think, not just WHAT to do
2. Decision trees → Agent can handle different scenarios
3. Self-checks → Agent validates its work
4. Progressive disclosure → Context-efficient

The Goldilocks Principle

"2-3 focused modules beat exhaustive documentation"

Keep it:

• - ✅ Complete enough to be useful
• ✅ Concise enough to fit in context
• ✅ Structured enough to guide reasoning

Example: Forging a Weather Skill

Step 1: Interpret

• - What: Fetch weather from wttr.in
• Triggers: "weather in [city]", "temperature", "forecast"
• Domain: Weather data

Step 2: Design

• - Complexity: Simple (just API calls)
• Structure: SKILL.md only
• Name: INLINECODE0

Step 3: Forge

CODEBLOCK11

Step 4-5: Test & Polish

• - Add more triggers ("sunny?", "rain?")
• Add error handling (wrong city, no network)
• Add presentation templates

Why This Works

Based on research (SkillsBench 2026), skills with reasoning frameworks perform better because they give agents a thinking structure, not just steps to follow.

Made with Skill Maker 🔨