Loading...
Loading...
Core philosophy for designing Claude Code skills - when to use skills vs agents, the knowledge test, and what makes skills valuable. Use when deciding component type or evaluating skill quality.
npx skill4agent add majesticlabs-dev/majestic-marketplace skill-design-philosophy| TEACH (Skill) | DO (Agent) |
|---|---|
| Coding conventions | Run linters and fix code |
| Workflow methodology | Execute multi-step processes |
| Framework patterns | Generate reports |
| Tool usage guidance | Fetch and analyze data |
| Skill | Why It Works |
|---|---|
| Coding style guidance - patterns Claude applies when writing code |
| Methodology knowledge - steps Claude follows for test-driven development |
| Framework patterns - conventions Claude uses for Stimulus controllers |
| Tool knowledge - how to use specific libraries and scripts |
| Skill | Why It Fails |
|---|---|
| "Code reviewer" | Does autonomous work - should be an agent |
| "Git helper" | Vague scope - what specifically does it teach? |
| "Best practices" | Too broad - not actionable |
| "Documentation generator" | Creates artifacts - should be an agent |
| Include | Exclude |
|---|---|
| Concrete patterns and conventions | Persona statements ("You are an expert...") |
| Specific templates and examples | Attribution ("Inspired by X...") |
| Decision criteria | Decorative quotes |
| Error handling guidance | ASCII art or box-drawing |
| Framework-specific idioms | Vague "best practices" |
"Does every line in this skill improve Claude's behavior?"