There are two hard things in computer science. I automated one of them.

"There are only two hard things in computer science: cache invalidation and naming things."

Every developer, eventually.

I can't help you with cache invalidation.

But naming things? I think I cracked that one. And the strangest proof is the name of the tool itself, which it chose on its own. Hang on for that part, because I didn't see it coming.

First, be honest about how you name things

Picture the last project you shipped.

You thought about the name for maybe ten minutes. You grabbed the first word that was free on npm and not too embarrassing. You told yourself you'd "fix it later."

You did not fix it later. It's still called new-project-2.

We all do this, and here's why it quietly costs us: a name is the one thing you can't refactor. You can rewrite the code, swap the stack, pivot the whole product. Rename it after launch and you pay for it in every dead link, every "wait, what's it called now," every bit of recognition you torched.

The companies that understand this don't wing it. They hire a naming studio that runs a careful, almost scientific process. Then they pay about the price of a used car for the result.

I wanted that process. I did not want that invoice. So I built it.

What you actually do

One line to install:

npx skills add bazingga08/nomira

Then, inside Claude Code, you just talk to it:

/nomira

a budgeting app for freelancers who hate spreadsheets.
calm and trustworthy, not a flashy finance-bro vibe.

It asks two or three sharp questions, the kind a good strategist asks before anyone is allowed to brainstorm a single word. Then it goes quiet and works. A minute later:

TOP 3 PICKS

1. Tendil      87  ·  strong
   "tend" + a soft landing. sounds like care, not accounting.
   trademark: clear in fintech. domain gettable.

2. Quillo      81  ·  strong
   the freelancer's pen, made friendly. easy to say and spell.
   trademark: one neighbor to check. handle is free.

3. Ledgerly    74  ·  caution
   clearest meaning of the three, but "ledger" is crowded.
   trademark: busy. you'd be fighting for it.

Every name comes with a reason, a sound note, and a risk flag.

Most tools hand you a haystack. This one hands you the needle, and shows you why it's the needle.

Why it beats a name generator

Most naming tools are a thesaurus with a loading spinner. They skip the part that actually decides whether a name is good.

Nomira runs the real pipeline, the same one a studio would bill you four figures for. In plain English:

• It thinks before it speaks. No names get made until it knows the one idea the name must carry. Boring step. It is the entire game.
• It brainstorms with the critic switched off. Hundreds of candidates, on purpose, so the weird-but-great ideas survive long enough to be noticed.
• Then it argues with itself. A strategist, a sound expert, and a trademark checker fight over the survivors. Whatever is still standing earned it.
• Then it scores cold. A real program rates each finalist and flags trademark and cross-language landmines, and it shows its math.

That last step is where I nearly shipped a lie.

The bug that taught me something real

My first scoring engine was a fraud, and it took me a while to catch it.

It would describe good rules ("penalize a name whose sound fights its meaning") and then quietly ignore them in the actual numbers. One time it ranked a name first while, in its own write-up, admitting that name failed its own checks. It was telling me what I wanted to hear and scoring on vibes underneath.

The fix was humbling: a flagged rule now has to mechanically drag the score down, not just appear in a paragraph.

If you have ever built anything that grades or ranks with an LLM, go look at it tonight. I would bet money it is doing the same thing.

The second lesson was quieter and bigger. The brief matters more than the generator. Garbage brief in, clever-but-hollow puns out. Sharp brief in, names you would actually register. It speeds up your taste. It does not replace it. You are still the one deciding.

So, did it really name itself?

Sort of, and the honest version is better than the clean one.

I pointed it at its own brief and took my thumb off the scale. Round one flopped, which honestly made me trust it more. I tightened the brief, ran it again, and it came back with three finalists:

Calliope   87.5   the muse of eloquence. gorgeous, but a mouthful, and crowded.
Nomira     83.5   from Latin "nomen," name. coined, complete, clean to own.
Atlas      n/a    instant gravitas, but trademark-crowded to death.

Calliope actually scored highest. I picked Nomira anyway.

Because a good name isn't just a number. It has to be ownable, easy to live with, yours. Which is the entire point of the tool: it gets you to a shortlist you'd be proud of, then you make the call. The scoreboard does the heavy lifting. The taste stays human.

A naming tool that can argue me down to three real choices is one I'll trust with yours.

Go name the thing you've been avoiding

You have one. The repo still called new-project-2. Give it thirty seconds:

npx skills add bazingga08/nomira

Then /nomira in Claude Code, and tell it what you're building.

Free. Open source (MIT). The repo even ships with the research it was built on, so you can read exactly why it makes the calls it makes:

https://github.com/bazingga08/nomira

Fair warning before you go: the trademark step is a smart pre-check, not a lawyer. Clear anything serious for real before you commit.

Now I'm genuinely curious how it does on yours. So here's a 30-second experiment with that new-project-2 repo:

1. npx skills add bazingga08/nomira
2. /nomira, then describe the thing
3. Drop two things in the comments: the name you settled on, and the name Nomira gave you.

I want to see how often the machine beats the human. My money is on a roughly even split, but I have been wrong about this tool before.