A Claude Code skill for naming products, SaaS tools, brands, and projects.

Metaphor-driven naming that produces memorable, meaningful names — and avoids AI slop.

When invoked, this skill guides Claude through a structured naming process:

1. Naming brief — establish what the product does, who it's for, what it should feel like
2. Metaphor exploration — map conceptual territories before brainstorming names
3. Candidate generation — produce names grounded in metaphor, not thesaurus surfing
4. Filtering — kill AI slop and anti-patterns
5. Evaluation — score and compare finalists with a weighted rubric
6. Availability checking — verify domains, handles, and package names
7. Decision — present top candidates with origin stories and trade-offs

Clone or copy the repo contents into your project's .claude/skills/naming/ directory:

# From your project root
mkdir -p .claude/skills
git clone https://github.com/glacierphonk/naming.git .claude/skills/naming

mkdir -p ~/.claude/skills
git clone https://github.com/glacierphonk/naming.git ~/.claude/skills/naming

In Claude Code:

/naming

Then describe what you need a name for. Claude will walk you through the full process.

You can also reference the skill naturally in conversation — describe your naming challenge and Claude will pull in the relevant reference files.

You don't need to read any of the reference files before using the skill. Just:

1. Type /naming
2. Describe what you're building in one sentence
3. Claude handles the rest — it loads the right references at each step

The 7-step process and 14 reference files are the depth layer. For a quick naming session, Claude compresses the process automatically. The reference files exist so you can dive deeper when needed.

New language and industry files welcome — see CONTRIBUTING.md for templates and required sections.

Names are compressed stories, not labels. The best names plant a concrete image that unfolds into understanding — what the product does, what it feels like, where it comes from.

This skill is opinionated:

• Metaphor over thesaurus. Don't search for synonyms of your product's category. Explore what else in the world works like your product.
• Real words over invented words. Real-word brand names have ~68.8% recall vs ~38.1% for invented names. The brain follows the path of least resistance.
• Story over sound. A name with a great origin story and average sound will outperform a name with perfect phonetics and no story.
• Kill AI slop. Suffixes like -ly, -ify, -able, meaningless portmanteaus, and thesaurus extraction produce polished-but-interchangeable names. This skill actively filters them out.

• Always loaded: Skill name and description (~2% of context budget)
• Loaded on invoke: SKILL.md (~180 lines, the process overview)
• Loaded on demand: Reference files load only when Claude reaches the relevant step. A simple naming task might only load 2-3 files; a thorough session loads 5-6
• Never auto-loaded: Language files, case-studies.md — only when explicitly relevant

Contributors: keep reference files focused. A 500-line file is fine; a 2,000-line file wastes context on content that may not be relevant.

PRs are checked by markdownlint and lychee (link checker) via GitHub Actions.

Common lint rules to watch:

• MD040 — fenced code blocks need a language tag (use text for plain text, bash for shell, markdown for markdown examples)
• MD001 — heading levels must increment by one (## → ###, not ## → ####)
• MD037 — no spaces inside emphasis markers. Use `___` (code backticks) for placeholders, not ___ (which looks like emphasis)

Run locally before pushing:

npx markdownlint-cli2 '**/*.md'

main is protected — all changes go through pull requests. Direct pushes are blocked.

PRs welcome. See CONTRIBUTING.md for conventions, file structure, and how to add language files or case studies.

MIT