name-clearance
Clear a product/brand name BEFORE buying a domain or branding anything. Runs same-industry collision, trademark signal, domain availability, and distinctiveness checks. Returns RED/YELLOW/GREEN verdic
SKILL.mdfactory/skills/name-clearance/SKILL.md
GitHubname-clearance
Why this exists (the Patently lesson). A product was named "Patently" and the domain
patently.legalwas bought — then we foundpatently.com: an 8-year-old, 1,500-user AI patent platform run by patent litigators. Identical name, identical field. The domain was available; the name was a landmine. A free domain proves nothing. Clear the name, not the domain.
When you activate
- A new product is being named (ideate → scaffold).
- Before any domain purchase — this is a hard gate in the go-live flow.
- A user proposes a brand name ("let's call it X").
- A rename is being considered.
What you produce
For each candidate, a verdict with evidence:
- 🟢 GREEN — no same-field collision, distinctive/arbitrary mark, domain obtainable. Safe to advance to a formal attorney knockout.
- 🟡 YELLOW — clear in-field but flags exist (crowded prefix, famous-brand crowding/SEO, common word, brand in an adjacent industry). Usable with caution + formal TM search.
- 🔴 RED — same/adjacent-industry collision, registered/strong common-law mark, or descriptive-and-taken. Kill it.
Plus a one-line recommendation and, for finalists, the explicit reminder: a paid attorney trademark knockout (USPTO + jurisdiction) before committing a cent.
Core principle — distinctiveness first
Trademark strength runs: fanciful > arbitrary > suggestive > descriptive > generic. Suggestive/descriptive names (Patently for patents, "IP Radar" for IP) are a double trap — legally weak and almost always already taken in-field. Bias hard toward coined or arbitrary names. In a crowded category (legal-tech, fintech, dev-tools) real dictionary words are mostly taken; a coined word is often the only reliably-clearable option.
But never assume a word is taken — check it. A registered .com is NOT a dead name;
many strong brands don't own their .com. What kills a name is a same-industry collision
or a live trademark, not a parked domain. Keep dictionary words in play through the check;
let the evidence (not a guess) eliminate them. (This skill is phase 4 of the
name-product pipeline — the clearance engine it calls.)
Protocol
Run cheap → expensive, fail fast. Stop at the first RED.
Distinctiveness triage. Place the name on the spectrum. If it's descriptive/suggestive of the product's own category, flag it immediately — it will be both weak and likely taken.
Same-industry collision (highest weight). Web-search the name against the product's field:
"<name>" <industry> company/"<name>" software startup/"<name>" <category-term>- Any company or product using the same or a confusingly similar name in the same or an adjacent industry → 🔴. (Adjacent counts: a trademark agency for an IP tool, film-IP-clearance for IP-clearance, fintech-decisioning for a risk tool.)
Trademark signal. Search the registries — USPTO TESS (https://tmsearch.uspto.gov), EUIPO eSearch, and the product's home jurisdiction — in the relevant Nice classes: 9 (software), 42 (SaaS/tech), 45 (legal services), plus any category-specific class. A live registration or strong common-law use (years of use, scale) → 🔴.
Famous-crowding / SEO. Is the root already owned in public consciousness by a famous brand, blog, or figure (e.g. "Patently-O" for patent content)? If the product will be invisible in search behind incumbents → 🟡 (downgrade), even absent a legal conflict.
Domain availability. Run the helper across acceptable TLDs:
bun ${HAMZAISH_ROOT:-$HOME/Claude/Hamzaish}/factory/skills/name-clearance/check-domains.ts <name> com ai legal app io co.comis ideal;.ai/.app/.legal/.ioare acceptable. The WHOIS verdict is a signal — confirm finalists at a registrar.Package-registry clearance (npm / PyPI / crates — for anything you'll publish). A 404 on the exact name means it's unregistered — NOT that you can publish it. Registries run a similarity / typosquat guard at publish time that rejects names too close to an existing package. (rotscan, 2026-06-21: the exact name 404'd as "available," then
npm publishreturnedE403 — too similar to existing 'rot-scan'.) So:- Search the registry for near-names, not just the exact one — hyphenated/un-hyphenated,
singular/plural, common char-swaps (
npm search <name>, or browse npmjs.com / pypi.org). - If a close look-alike exists, expect rejection — plan to scope it (
@you/name, npm's own suggested fix) or pick a more distinctive name up front. npm publish --dry-rundoes not run the similarity check; only a real publish does. Scoping (@you/name) sidesteps it entirely.
- Search the registry for near-names, not just the exact one — hyphenated/un-hyphenated,
singular/plural, common char-swaps (
Linguistic / social sanity. Unfortunate meanings in other languages, hard to spell/say, handle availability (X / GitHub / LinkedIn).
Verdict & recommendation. Summarize each candidate as 🟢/🟡/🔴 with the evidence (links). Recommend the strongest GREEN. For any finalist, state plainly: get a paid attorney knockout before committing — this skill de-risks, it is not legal advice.
Output format
A compact table: Name | Verdict | Same-field collision | Domain | Note, then a one-paragraph
recommendation. Cite the URLs that drove each verdict.
Notes
- Run on a batch of candidates, not one — most will fail; generate 8–12 to find 1–2 GREENs.
- Re-run the whole protocol after any rename; don't trust a half-check.
- This skill is a required pre-step before the domain-buy stage of
factory/playbooks/ai-native-2026/go-live-provisioning.md.