new-skill

Claude Code skill that scaffolds a new skill in seconds.
Answer 8 questions, get a fully structured skill repo with all conventions baked in.

---

Install

git clone https://github.com/ibalasite/new-skill.git ~/.claude/skills/new-skill
~/.claude/skills/new-skill/setup install

Restart Claude Code after install.

---

Usage

/new-skill

Claude asks 8 questions, then generates a ready-to-code skill directory at ~/projects/{skill-name}/.

---

What the 8 questions capture

#	Question	Why it matters
1	Skill name (kebab-case)	Becomes repo name, ~/.claude/skills/{name}/, slash command
2	One-line purpose	README headline, skill.md description
3	Scope boundary — what it does NOT do	Prevents runtime pollution of source repo
4	Data model — where state lives	Lazy-create rules for hook scripts
5	Hook types + behavior	Stop / PostToolUse / SessionStart registration
6	Sub-skills list	What gets deployed to ~/.claude/skills/ on install
7	External dependencies	Prereq checks in setup script
8	Feasibility risks	Known limitations section in CLAUDE.md

---

What you get

~/projects/{skill-name}/
├── CLAUDE.md       ← full spec, filled with your answers
├── README.md       ← user-facing docs
├── progress.md     ← standard + skill-specific implementation steps
├── skill.md        ← /{skill-name} entrypoint
├── setup           ← install / update / uninstall
├── bin/            ← ready for hook scripts and shared libs
└── skills/
    └── {skill-name}-upgrade/
        └── skill.md

Then follow progress.md — Step 1 is already DONE.

---

Conventions every generated skill inherits

1. Source repo = source only. No runtime ops during development.
2. ~/.claude/skills/{name}/ is the runtime. Clone = install.
3. Sub-skills via skills/. Deployed on install AND update.
4. Project data = lazy. Hooks create {project}/.{name}/ on first run.
5. Uninstall preserves data. Never deletes user project directories.

---

Update

/new-skill-upgrade

Or: ~/.claude/skills/new-skill/setup update

Uninstall

~/.claude/skills/new-skill/setup uninstall