agentleFS
Sign inStart free
JPInertJPInert/claude-code-skills/skills/domainhunt/SKILL.md

domainhunt

Skill0 starsChanged 39 days agoMITView on GitHub

Check whether domains are available and sweep thousands of candidate names at once, without being fooled by a rate-limited checker. Use when the user says /domainhunt, "find a domain", "is that domain taken", "check if X.com is free", "what names are available", "sweep for names", "what else is available", "how much is that domain", "is that TLD expensive", or asks you to name a product or service. Also use before telling the user a domain is available or that a name is a good one.

  • Sends data out

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 3 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Test a known positive BEFORE and AFTER every batch: Before-only is not enough: the throttle arrives *during* the run.

    line 18
  2. Distinctive:

    If the after-test fails, the whole batch is void.

    line 23
  3. Distinctive:

    Never use it. Go to the registry directly.

    line 35
  4. Very specific:

    Keep allchecked.txt and a short list of dead shapes per project, so a later session does not re-sweep them.

    line 110

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md
<!-- needs: bash, curl, dig (dnsutils/bind-utils), python3 for run_bank.sh. No accounts, no hardware. --><!-- pricing lookups use Porkbun's public, unauthenticated pricing API. Registering is never automated. -->

Domain hunting

Two jobs, and the second is the one that goes wrong: find candidates, and not lie about whether they are free. Every instrument here fails toward false available when throttled, so a sweep that reports thousands of free names is the expected output of a broken run.

Scripts live in scripts/ next to this file. They need only curl, dig and bash.

The instrument rule: this is the whole skill

Test a known positive BEFORE and AFTER every batch. Before-only is not enough: the throttle arrives during the run.

  • ./scripts/w.sh google.com must say TAKEN; ./scripts/w.sh zzqxjfkdlqp123.com must say FREE.
  • ./scripts/ns2.sh google.com must print nothing; the nonsense name must print.
  • If the after-test fails, the whole batch is void. Do not salvage part of it.

What this caught, so it is not re-learned at cost:

  • dig at -P 40 over 95,000 names: the resolver throttled and returned empty for everything, which reads as no nameservers = available. All 95k were garbage.
  • WHOIS run concurrently with a big dig sweep returned false No match. A 40-name random sample of "free" survivors was 40/40 actually registered: fluids.com, aussie.com, bourse.com. Serial re-checks said TAKEN.
  • Verisign RDAP at -P 5 returned HTTP 000 for every name including google.com. A registry block is indistinguishable from an outage.
  • rdap.org (the bootstrap aggregator) returned 404 for everything, including github.io. Never use it. Go to the registry directly.

Pipeline by size

namesmethod
< 500WHOIS direct, -P 3, no prefilter. ~30 s. Safest
500 – 25,000dig NS prefilter -P 20 → WHOIS -P 3 on survivors
> 25,000don't. Split it. The prefilter throttles and the run is void

./scripts/run_bank.sh <wordsfile> <prefix> does the whole middle row: crosses the words against a join list (JOIN="fix hop desk" in the environment; your product's short stems) in both orders, dedupes against allchecked.txt, prefilters, confirms, and self-tests at both ends. It exits non-zero and prints VOID when a self-test fails, so a script can gate on it. Keep a cumulative allchecked.txt of every candidate ever generated: re-learning that fixatom is taken is pure waste.

The instruments

  • ns2.sh <domain>: prefilter. Prints the domain only if no nameservers on 1.1.1.1, then 8.8.8.8, then 9.9.9.9. An unregistered domain cannot have NS, so this has no false negatives: only false positives, which the confirm step catches.
  • w.sh <domain>: WHOIS port 43 at whois.verisign-grs.com. Bash's /dev/tcp does not reliably resolve hostnames, so the script resolves the IP first with getent and connects to that. .com/.net only. No match for = FREE, Domain Name: = TAKEN, anything else = retry then UNKNOWN.
  • conf.sh <domain>: RDAP at Verisign. 404 = FREE, 200 = TAKEN. Second instrument; blocks above -P 4.
  • multi.sh <domain>: other TLDs, registry-direct endpoints. Refresh the table from https://data.iana.org/rdap/dns.json. .us = whois whois.nic.us, answers No Data Found. .co has no working RDAP endpoint: report it as unchecked, never guess.

Two instruments before you recommend anything. RDAP and WHOIS are independent; make both say FREE, serially, on the day the user might buy.

The five acceptance checks: run them BEFORE presenting a name

A free domain is not a usable name. In an industry scammers impersonate (remote IT support, for example) a bad name is a liability.

  1. Reversal. For fixX, check Xfix. Is it registered? Does it serve anything? curl -sL it and read the <title>.
  2. Live product with that name. One WebSearch for the exact compound. This caught WarpFix (live, selling, $12/mo, warpfix.org), Warpgate (7.4k-star SSH/RDP bastion, the same category), VoidFix (live IT company), CalmPC (live PC health checker), SteadyFix (phone repair shop), Hopgate (indie-games platform), Parsec (remote desktop). All would have shipped without this check.
  3. .net / .org flanks: cheap to know, and owning the pair kills the reversal problem.
  4. Typo-adjacency to a competitor. Edit distance ≤2 to a competitor's name = reject. For remote desktop that list was anydesk, teamviewer, splashtop, hoptodesk, rustdesk, logmein, gotomypc, simplehelp, getscreen, quickassist. jumptodesk/poptodesk are free because they are HopToDesk lures.
  5. Is either half a household brand? This is the one that kills good-looking names: fixcalm→Calm the app, fixcomet→Comet the scouring powder, hereassist→Here2Assist® (Takeda). Say the name out loud as "go to ___ dot com" and listen for the other brand.

Pricing: measure, never estimate

Porkbun's pricing API is public and unauthenticated: curl -s -X POST https://api.porkbun.com/api/json/v3/pricing/get

Measured 2026-09-22: .com $11.08 reg and renew · .io $28.12 then $51.80/yr · .co $31.20 renew · .tech $50.98 · .support $22.14 · .help/.live $26.26 · .app $14.93 · .dev $12.87 · .net $12.52 · .org $11.84 · .cc $8.55 · .link $7.72 · .us $7.00 · .fyi $5.66.

.com undercuts almost every alternative TLD's renewal, so a longer .com beats a short anything-else on both price and trust. Quote the renewal, not the first year. Per-domain premium pricing needs a registrar API key: say so rather than guessing.

.app, .dev and .page are HSTS-preloaded (verify at https://hstspreload.org/api/v2/status?domain=app): browsers refuse plain HTTP and there is no click-through on a bad cert: which removes the safety net on a self-signed migration.

Keep a record of what is exhausted

Keep allchecked.txt and a short list of dead shapes per project, so a later session does not re-sweep them. From one 52,000-name, 18-sweep hunt for a remote-support product name (September 2026), as an example of what that list looks like: single English words of 4–7 letters, two words ≤8 chars with any meaning, stem+suffix (-ly -ify -io …), letter or number + word (ifix, 4fix), and 8–9-letter single words (2 of 63 free) were all dead. Still alive: fix + an uncommon concrete noun at 8–10 chars, verb + fault word, and 3-word phrases in the user's voice (944 of 1,243 free in one sweep).

Never buy on an inferred go-ahead

Registering costs money and cannot be undone. Picking a favourite is not authorisation: the user has to say buy.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free