Domain Scout

Domain Scout

Turn a concept into a ranked shortlist of available, brandable, buyable domains. You (the model) are the creative name generator; scripts/scout.mjs is the mechanical availability checker. The output is a small ranked shortlist with the reasoning and a recommended pick, the way a good namer would hand it over.

The core discipline: availability ≠ good, and "feels open" ≠ open. A name is only a candidate once whois says it's free AND no established company already owns the brand. Check, never eyeball (the recurring lesson: a long-registered .com signals an existing company, even if a variant is free).

The run

1. Frame the concept in one line: what it is, who it's for, the feeling. Note any constraints (must be .com? short? contain a keyword? avoid a word?).
2. Generate candidates across angles (below), 15–30 names. Widen the stock before narrowing.
3. Check availability with the script (whois across TLDs + variant patterns).
4. Collision screen the survivors: a quick web search per finalist for an existing company using the name (esp. same/adjacent category). Flag conflicts.
5. Rank by the brandability rubric and present the shortlist: domain, available?, why it fits, any collision flag, and the one you'd pick with the reason.

Name-generation angles (widen the stock)

Generate from several of these, not one, or you get monotone output:

• Descriptive / literal — says what it does (vendormatch, officesourcing). Clear, often taken, low distinctiveness.
• Evocative / metaphor — a concrete image for the benefit (beacon = guidance; northstar). Memorable, on-theme.
• Invented / portmanteau — fuse two relevant words (vendra, sourcely, qualifi). Brandable, often available.
• Real word, angled — an ordinary word used slant (slate, vouch, roster). Premium feel, .com usually taken → needs a variant.
• Compound — two short words (workbid, sitelaunch, leadloop). Reads instantly.
• Latin / Greek / foreign root — for gravitas or when English is exhausted (certus, civ, mano).
• Industry jargon as a name — the term insiders already use ("churn-saver" for a retention play → Churnsaver). Self-explanatory to the target market; strong signal.
• Prefix / suffix patterns — when the bare word is taken: get___, use___, try___, join___, ___hq, ___app, ___labs, go___. ___hq and get___ read as real startups; try___ implies beta.

In a hot / picked-clean namespace (anything agent-* or AI-adjacent in 2026), the descriptive space is gone — checking 40 agent* / *ai names can surface zero available. Lean hard on invented / coined (the Perplexity / Anthropic / Mistral pattern funded startups actually use) and check widely; don't burn the run on descriptive names that are all taken.

TLD strategy

.com is the strong default — most trusted, the one people type and say by reflex. Prioritize it hard: treat .ai / .co / .io / .org as fallbacks that need a reason, not peers. A clean available .com should outrank a nicer name stuck on a non-.com, and a prefixed .com (get___, ___hq, try___) usually beats dropping the TLD entirely.

1. .com — default; if the bare .com is taken, try a prefix/suffix .com before leaving .com.
2. .co — closest substitute, pricier, slightly weaker.
3. .io / .dev / .app / .ai — acceptable ONLY for dev/AI/technical audiences with a reason; .ai reads on-trend but costs more and isn't .com.
4. .xyz / others — only with a strong reason.

So: check and rank .com first, always. Only present a non-.com as the top pick if there's no viable .com AND the audience genuinely accepts the alternative. When in doubt, .com.

Brandability rubric (score each survivor)

• Length — shorter is better; ≤ 2 syllables / ≤ 10 chars is ideal.
• Say-it / spell-it — can someone hear it once and type it right? No silent letters, no ambiguous spellings, no numbers, no hyphens (hard bans: numbers, hyphens).
• Meaning fit — does it evoke the concept or the benefit?
• .com availability — heavily weight a clean .com (or an acceptable prefixed .com).
• Collision-free — no established company owning the brand (see collision screen).
• Distinctiveness — would it get lost among lookalikes? (competitive-exclusion: different enough to own.)

Score = brandability × availability × fit × collision-free. Present the top 3–5, not all.

Availability check (the script)

# check specific full domains
node scripts/scout.mjs check leadloophq.com useleadloop.com leadloop.co

# check a base name across default TLDs + common prefixes
node scripts/scout.mjs base leadloop
node scripts/scout.mjs base leadloop --tlds com,co,io,ai --prefixes get,use,try,join,hq

Multi-signal, no API key: DNS nameservers (registered domains always have NS → definitive taken), then RDAP (structured JSON; clean 404 = available, 200 = taken), then whois as last resort. This is reliable across TLDs including .ai — plain whois lies on .ai (it false-positived a registered domain as available), which is why DNS+RDAP lead. Output per domain: AVAILABLE / taken / unknown (re-check). It's a fast filter; confirm at purchase, never treat as a registry guarantee.

Collision screen (don't skip)

For each finalist, a 2-minute check before recommending: WebSearch "<name>" company (and the bare .com if taken — a long-registered .com usually means a real company holds the brand). If an established company uses the exact name, especially in the same or an adjacent category, flag it. A different-category collision is usually fine as a working name but worth a trademark gut-check before committing.

Output shape

A short ranked list. For each: domain · available? · price-tier if known · one-line why · collision flag. End with the single pick and the reason, the way a namer hands off a recommendation, not a data dump.

Angles to grow (this skill is meant to expand)

Room for more lenses over time: registrar price lookup (premium names cost more), expiring/dropped domain hunting, thesaurus/semantic expansion of the concept, phonetic neighbors, two-word availability sweeps, social-handle availability (matching @handle), and AI bulk-generation with a scored filter. Add each as a new mode or script rather than bloating the core run.

Whatever gets added, the core discipline holds: check, never eyeball; confirm at purchase.