Claude Skill Builder: generate SKILL.md scaffolds
Generate production-ready Claude Skill SKILL.md files with YAML frontmatter, instructions, examples and validation. Works with Claude Code, Claude Desktop, the Claude API (Agent Skills), and the Claude Agent SDK. Live preview + download. Browser-only.
What you'll build with Claude Skills
Code reviewers
A skill that scans pending git diffs for security bugs, leftover console.logs, missing tests, etc. — invoked automatically when you say "review this PR".
Domain-specific Q&A
Bundle company docs / API reference / changelog so Claude answers in your house style — no RAG infrastructure needed.
Build-system helpers
"Rebuild and deploy to staging" — a skill that knows your Makefile, env vars, and deploy script. Saves typing every detail.
Schema/data helpers
A skill that knows your database schema and writes correct SQL on request. Update the SKILL.md when the schema changes.
Debugging playbooks
"How do we usually fix this error?" — capture team tribal knowledge so Claude references the same runbook every time.
Content formatters
A skill that converts meeting notes → blog post, Slack thread → docs page, or transcript → action items in your preferred format.
How to build a Claude Skill
Pick a template or start blank
The 6 templates above cover most common cases. Use Empty if your skill is unusual.
Name it (kebab-case)
Skill name must be kebab-case, 3–60 chars, unique within a skills folder. The name is also the folder name.
Write the description carefully
The description is what Claude reads to decide whether to load your skill. Be specific about when to trigger it. Vague descriptions get over- or under-triggered.
Add instructions + examples
Step-by-step instructions are the body. 1-3 good examples teach Claude the exact output format faster than 1000 words of prose.
Download and install
Save SKILL.md to ~/.claude/skills/{name}/SKILL.md (Claude Code, user-wide) or .claude/skills/{name}/SKILL.md (per-project). Restart Claude Code if it was already running.
About Claude Skills
Claude Skills are reusable capability bundles: a folder with a SKILL.md instruction file and any supporting files (scripts, references, examples). When you describe a task that matches a skill's description, Claude auto-loads the SKILL.md into context — no need to re-type instructions every session.
Supported runtimes
- Claude Code (CLI + IDE extensions) — looks in
~/.claude/skills/{name}/(user-wide) and.claude/skills/{name}/(per-project) - Claude Desktop (Mac/Windows app) — Settings → Capabilities → Skills folder
- Claude API (Agent Skills) — pass via the agent skills array on the messages.create call
- Claude Agent SDK — register via the agent_skills parameter when building an agent
Anatomy of a SKILL.md
---
name: my-skill
description: Concise sentence describing exactly when to trigger.
allowed-tools: Read, Bash, Grep # optional, Claude Code only
---
# Skill Title
## When to use
Specific trigger conditions. Mention example user requests.
## Instructions
1. Step-by-step what Claude should do.
2. Each step clear and atomic.
## Examples
**User**: Example request
**You**: Example response
## Output format
Describe expected output structure.
## Notes
Edge cases, constraints, error handling.What makes a good skill
- Specific description. "Helps with code" triggers always — useless. "Reviews pending git diffs for security bugs and missing tests" triggers correctly.
- Concrete instructions. Numbered steps, not prose. Each step does one thing.
- 2-3 examples. Teaches output format faster than any prose description.
- Failure modes. Tell Claude what to do when the input is wrong or tools are missing.
- Small focus. One skill = one capability. Splitting is better than combining.
YAML frontmatter rules
name: kebab-case, 3–60 chars. description: 1-3 sentences, no markdown. allowed-tools: comma-separated list (Claude Code only — Read, Edit, Bash, Grep, Glob, etc.). The triple-dash separators must be on their own lines.
Privacy
This builder is pure JavaScript — your skill content is generated entirely in your browser. Nothing is sent to a server, including the download.
Claude Skill Builder — FAQ
A reusable capability bundle: a folder with SKILL.md (instructions + YAML frontmatter) and any supporting files. Claude auto-loads the SKILL.md when its description matches the user's request. Works in Claude Code, Claude Desktop, the API and the Agent SDK.
YAML frontmatter (between ---) with name and description. Body is plain Markdown — typically ## When to use, ## Instructions, ## Examples, ## Output format. name must be kebab-case and unique in its skills folder.
Claude Code: ~/.claude/skills/{name}/SKILL.md (user-wide) or .claude/skills/{name}/SKILL.md (per-project). Claude Desktop: Settings → Capabilities → Skills folder. Claude API: pass via the agent skills parameter. Always one folder per skill.
Very specific. The description is what Claude reads to decide whether to load this skill. Vague descriptions get over-triggered ("helps with code"). Specific descriptions get triggered only when relevant ("Reviews pending git changes for security bugs at low/medium/high effort").
Yes. Put helper scripts in the skill folder alongside SKILL.md. Reference them in your Instructions section with relative paths. Claude can read, execute, or include them in context.