One Star

I ran the numbers on my own GitHub account this week, the way I'd audit a database. Thirty-six public repositories. Three published npm packages. Multiple production systems with real users. And exactly one star from a stranger — a person in Franklin, Pennsylvania who found FleetCrown on June 19 and clicked. My other star turned out to be me, starring my own repo. My one visible follower follows fourteen thousand people; that ratio is the signature of a bot.

So: one human, ever, has found this work and cared.

Most people would keep that chart private. I'm publishing it, because it's the honest baseline of an experiment I want witnessed from the start — and because the diagnosis matters more than the embarrassment: the work has never been the problem. Nobody has ever been shown it. Zero launches. Zero posts. Zero submissions to a newsletter, a subreddit, anything. One star is not what failure looks like; it's what silence looks like.

What the silence has been hiding

This is the part where I'm supposed to be modest, but modesty and dishonesty aren't the same thing, so here is what exists as of tonight — every claim is a link:

All three packages publish without any npm token existing anywhere — OIDC trusted publishing, releases are just merged version bumps. That pipeline was built two nights ago, by the thing I actually want to show you.

The thing I actually want to show you

I'm one person. I don't have a team. What I have is a fleet of AI agents and a harness of automation that does the judging a team would do — and the harness, not any single app, is the most interesting thing I own.

Two nights ago this loop ran a whole campaign while I mostly slept: three packages migrated to tokenless publishing, bip-kit extracted, published, and adopted by two production apps — and one of the merges was written by two different agents that had never been told about each other. One prepared the dependency migration; the other found the pushed branch, stacked the docs and dogfood essays on top, and opened the pull request itself. The sweep merged it. The reconciler deployed it. The health probes proved it live. My total contribution was wanting it, judging it, and four taps of a security key.

I've now extracted that harness into a public repo: crewless. It's v0 and it says so — the scripts are the canonical copies from the live fleet, with the outage that shaped each one documented in its header. The auto-merge sweep alone carries the scars of a GitHub Actions incident that stranded eleven pull requests for fourteen hours, and of being copy-pasted into twenty-two repos where it drifted into eight incompatible versions before I re-unified it. Software with a memoir.

Why I believe this is bigger than my account

Solo builders used to be rare, and their products mostly deserved the obscurity. That constraint is gone — not because builders got smarter, but because the bottleneck moved from hands to judgment. One person who encodes their judgment into gates can direct more correct work than they could ever type. Today that's software. The same loop, pointed at robots, builds you a pool while you're on vacation — I wrote about that arc in FleetCrown's essays, and I don't think it's far-fetched, just early.

Building in public is what this new kind of builder does instead of marketing: the audit log — what was asked, what the fleet did, how it was proven — published as it happens. This article is itself an entry in that log, written the night the experiment started, on a platform I built, rendered by a package (bip-kit) this campaign shipped.

The experiment, stated so you can hold me to it

Starting numbers, August 22, 2026: 1 organic star. 0 real followers. 0 launches.

The plan, in public:

1. One anchor: crewless grows from honest extraction to something you can adopt in an afternoon.
2. One launch per week, smallest first — ai-forms, then bip-kit, then crewless — Show HN, newsletters, the relevant communities. No growth hacks, no bought stars, ever; a faked chart would poison the only asset this experiment has, which is that every number in it is real.
3. One essay per week, here on OrangeCat, documenting what actually happened — including the launches that flop.
4. Every stranger answered within a day. At one star, each new person is a doubling.

If this works, you'll be able to see exactly how, week by week, from a true zero. If it doesn't, that log will exist too, and it will be just as specific.

The chart above is the before picture. Star the repo if you want to be in the after one — you'd roughly double my audience.