skill-creator-master

Production-grade skill creation framework for AI agents.

Build, validate, and publish professional skills for Claude Code, Cursor, Gemini CLI, and other AI agents. Battle-tested from real skill development on skills.sh.

---

Install

npx skills add https://github.com/emrekent/skill-creator-master --skill skill-creator-master

What It Does

When you say "Create a new skill" or "Help me build a skill", this skill guides you through:

1. UNDERSTAND → Gather concrete usage examples
2. PLAN       → Identify scripts, references, assets needed
3. INIT       → Generate skill folder structure (automated)
4. BUILD      → Write SKILL.md + bundled resources
5. VALIDATE   → Run validation suite (automated)
6. PUBLISH    → Push to GitHub → verify discovery → first public install → live on skills.sh

---

What's Inside

skill-creator-master/
├── SKILL.md                    Core skill guide (under 500 lines)
├── scripts/
│   ├── init_skill.sh           Generates new skill folders automatically
│   ├── validate_skill.sh       Validation before publishing
│   └── verify_publish.sh       Checks CLI discovery and skills.sh listing state
└── references/
    ├── structure-guide.md      Skill anatomy, YAML rules, file organization
    ├── lessons-learned.md      Real-world case study from translator-ai
    ├── anti-patterns.md        10 common mistakes and how to avoid them
    └── checklist.md            Pre-publish validation checklist

---

Included Scripts

Init Script — Generate a New Skill

bash scripts/init_skill.sh <skill-name> <output-directory>

Creates a complete skill folder with SKILL.md template, scripts/, references/, and assets/ directories.

Validate Script — Check Before Publishing

bash scripts/validate_skill.sh <path/to/skill-folder>

Runs automated checks for:

• YAML frontmatter validation (name, description)
• SKILL.md line count (must be under 500)
• Placeholder detection
• Auxiliary file detection (warns about extra repo docs)
• Phantom reference detection
• Script executability
• Naming consistency

Publish Verification Script — Check Real Listing Status

bash scripts/verify_publish.sh https://github.com/<username>/<repo> <skill-name>

Checks:

• The upstream skills CLI can discover the repo from GitHub
• The specific skill name is visible to the CLI
• The skills.sh search API contains the skill
• The skills.sh/<owner>/<repo>/<skill> page resolves

If the page is still missing, the script explains the usual reason: skills.sh listing is driven by successful public installs through the skills CLI, not by a GitHub push alone.

How skills.sh Publishing Actually Works

1. Push the skill to a public GitHub repository.
2. Verify discoverability with npx skills add <repo> --list.
3. Perform at least one real public install with npx skills add <repo> --skill <skill-name>.
4. Re-check https://skills.sh/<owner>/<repo>/<skill-name>.

Important:

• --list verifies structure and frontmatter, but it does not publish.
• Local-path installs do not create skills.sh listings.
• skills.sh search and page generation are driven by anonymous install telemetry from the skills CLI, so indexing can lag after the first public install.

---

Key Principles

• Under 500 lines — SKILL.md stays lean. Details go to references/
• Description is the trigger — Keywords in description determine when the skill activates
• Progressive disclosure — Metadata always loaded, body on trigger, references on demand
• Working scripts only — No placeholder code. Everything runs.
• NOT clause required — Tell the AI what NOT to use the skill for

---

Battle-Tested

Built from real experience creating translator-ai — a professional translation skill with 7 language pairs, 100+ examples, and 98%+ quality assurance. Every lesson learned is documented in references/lessons-learned.md.

---

About the Creator

Built by Emre Kent — Systems architect & builder. Former math teacher turned automation designer.

I design growth engines, AI workflows, and automation systems. I build CRM architectures, AI agents, n8n automation systems, and personal operating systems that compound.

For discussions on AI agents, skills architecture, and automation systems — connect on LinkedIn.

---

Questions & Community

Have questions? Found a bug? Got a suggestion?

Open a Discussion or Issue.

---

License

Open source. Free to use, adapt, and build on.