Documentation

SkillSmith walks you through a 4-step wizard to create production-ready Claude Code skills:

The 4-Step Wizard

1. Context — Paste your profile, describe your role, or upload files. The AI analyzes your background and workflow.
2. Ideas — The AI suggests 2-3 tailored skill ideas based on your context. Each has a complexity tier badge. Pick one or write your own brief.
3. Draft — Your skill is generated as a complete package: SKILL.md plus companion files. Edit, refine, test, and iterate.
4. Download — Download the finished skill as a .zip, or share it to the public gallery.

Profile Parsing

Paste a link to your LinkedIn, GitHub profile, portfolio site, booking page, or resume text. The AI extracts signals like your tech stack, job title, responsibilities, and projects to suggest highly-specific skills you would actually use.

Complexity Tiers

Each skill idea is tagged with a complexity tier:

• Simple — SKILL.md only. Best for straightforward automations.
• Moderate — SKILL.md + 1-2 companion files. Good for workflows that need templates or reference data.
• Complex — SKILL.md + 3+ companion files. For comprehensive workflows with scripts, examples, and reference docs.

Click the tier badge on any idea to override it before generating.

Custom Briefs

Skip the AI suggestions and write your own skill brief. Describe what you want the skill to do, what tools it should use, and any specific behavior. The generator will follow your instructions.

Claude Code skills are Markdown files with YAML frontmatter. The file is always named SKILL.md and lives inside a skill folder.

SKILL.md Structure

---
name: my-skill-name
description: What this skill does and when to use it
version: "1.0"
author: Your Name
tags:
  - automation
  - testing
---

# My Skill Name

Instructions for Claude Code go here.
The body supports full Markdown.

Progressive Disclosure

Well-structured skills use 3 levels of detail to keep the context window efficient:

• Level 1 — Core instructions in the SKILL.md body (always loaded)
• Level 2 — Companion files referenced with "Read scripts/helper.sh for details" (loaded on demand)
• Level 3 — Dynamic context injected at runtime via !command

Companion Files

Skills can include additional files organized in standard folders:

• scripts/ — Shell scripts, Python helpers, automation tools
• references/ — API docs, style guides, coding standards
• examples/ — Sample inputs, expected outputs, templates
• assets/ — Config files, data, or other static resources

String Substitution

Use $ARGUMENTS in your skill body to receive the full argument string passed by the user. Use $1, $2, etc. for positional arguments.

Dynamic Context Injection

Prefix a shell command with ! in the frontmatter or body to inject its output at load time. For example, `!git log --oneline -10` injects the latest 10 commits into the skill context.

The Draft step gives you a full editing suite with multiple tabs:

Edit Tab

Direct editing of SKILL.md content with syntax highlighting. Changes are validated in real-time: YAML frontmatter parsing, kebab-case name format, description length limits.

Preview Tab

Rendered Markdown preview of your skill, exactly as it will appear when installed.

Files Tab

Manage companion files: view, edit, add, or remove files in scripts/, references/, examples/, and assets/ folders.

AI Refine

Describe changes in natural language and the AI rewrites your skill. Previous state is saved automatically, enabling the diff view and one-click revert.

Frontmatter Configurator

A sidebar UI with form fields for every official Claude Code skill frontmatter field. Two-way bound: changes in the form update the SKILL.md and vice versa.

Description Optimizer

AI generates 10 evaluation queries (5 should-trigger, 5 near-miss should-not-trigger), scores your description accuracy, and suggests improved wording. Helps ensure Claude Code loads your skill at the right time.

Test Tab

The Test tab simulates how Claude Code would respond with your skill loaded. Type a user message and see streaming output as if Claude Code were running. Limited to 5 tests per session.

Quality Analysis

Click "Analyze Quality" to get a structured 1-10 score with detailed feedback: strengths, weaknesses, and specific suggestions. Suggestion chips are clickable and auto-fill the refine prompt for quick iteration.

Before/After Diff View

The Changes tab shows a GitHub-style green/red diff after each refinement. See exactly what changed at a glance.

Revert to Previous

Not happy with a refinement? One click reverts to the previous version. The diff view makes it easy to decide.

Zip Structure

Downloaded skills are packaged as a .zip with this structure:

my-skill-name/
  SKILL.md
  scripts/
    helper.sh
  references/
    api-docs.md
  examples/
    sample-input.json

Installing in Claude Code

Unzip the skill folder into your Claude Code skills directory. There are two locations:

• Project-level — .claude/skills/my-skill-name/ in your project root. Available only in that project.
• User-level — ~/.claude/skills/my-skill-name/ in your home directory. Available in all projects.

Once installed, invoke your skill in Claude Code with /my-skill-name or let Claude Code auto-load it based on the description.

Submit Flow

After generating a skill, click "Share to Gallery" on the Download step. Add optional author attribution, then submit.

AI Auto-Review

Every submission is automatically reviewed by Claude Sonnet 4.6, which scores it for safety, quality, usefulness, and originality. Skills meeting the threshold are auto-approved; borderline cases are flagged for manual review.

Gallery Listing

Approved skills appear in the public gallery. Each listing shows a rendered preview, metadata sidebar, tags, and a download button.

Download Counts

Each skill tracks how many times it has been downloaded. Popular skills surface higher in search results.

Complete reference for all supported SKILL.md frontmatter fields:

For the full Claude Code skill specification, see the official Anthropic docs.