Six-Step Skill Building Framework

A structured process for creating a Claude Code skill from scratch. Nate Herk presents this as the standard before every live skill build, ensuring the skill has clear intent, process, guardrails, and a self-improvement loop.

The six steps

1. Name and trigger — what you type to invoke the skill, and when it should fire automatically ^[src-013]

2. One-sentence goal — a single declarative sentence describing what success looks like ^[src-013]

3. Step-by-step process — the numbered procedure the skill follows, written in imperative language ^[src-013]

4. Reference files — any scripts, templates, or external data the skill needs, stored in the skill folder ^[src-013]

5. Guardrails and rules — what the skill must never do, edge-case handling, output quality constraints ^[src-013]

6. Self-improvement feedback loop — instructions for the skill to ask for feedback after each run and update itself ^[src-013]

Key points

• Steps 1–3 define the contract; steps 4–6 make it production-grade ^[src-013]
• Without step 6, the skill never improves; the Skill Feedback Cycle turns this into a continuous quality loop ^[src-013]
• Skills should stay under ~500 lines (Anthropic guidance); use reference files for overflow ^[src-013]
• A skill written by following this framework takes 5–10 minutes; a skill written informally takes 20+ iterations to stabilise ^[src-013]

Related entities

• Claude Code Skills — the entity this framework builds
• Claude Code — the runtime

Related concepts

• Skill Feedback Cycle — the iterative quality loop in step 6
• Progressive Context Loading — why reference files in step 4 load on demand
• Global Vs Project Skills — where the resulting skill file lives
• Four Cs Aios — skills are the "Capabilities" layer (C3)

Source references

• ^[src-013] Nate Herk — "Build & Sell Claude Code Operating Systems (2+ Hour Course)" (2026-05-01)