I Sold 500 Copies of My First Claude Skill. Most of My Decisions Were Wrong

My first skill was a meeting-notes summarizer. Nothing exotic. You feed it a raw transcript and it hands back three things: the decisions that actually got made, the action items, and who owns each one. I built it because I was tired of scrolling through 4,000-word transcript dumps at 6pm trying to remember what I'd agreed to. I listed it on Skillmint on a Tuesday afternoon, mostly as an experiment. It has now sold 512 copies.

I'm genuinely happy about that. I also made nearly every mistake a first-time seller can make, and I made most of them confidently. So here's the unflattering version — the one I'd actually want to read.

I priced it like I was apologizing

I launched at $7. My reasoning, if you can call it that, was "it's a simple skill, people will balk at anything more." What actually happened is that $7 reads as "this is probably junk someone threw together in an afternoon." Which, fine, I did throw it together in an afternoon — but the buyer doesn't know that, and the price was doing the talking before anyone read a word.

Four weeks in, on a dare from a friend who sells online courses, I raised it to $19. I braced for sales to crater. They went up. Not flat — up, by something like 40% week over week. Price is a quality signal that arrives before anyone has tried the thing. A low number doesn't say "great deal," it says "low stakes, low effort, probably broken."

If you've built something that reliably saves a person an hour, $7 is an insult to your own work. Charge like that hour is worth something, because to the buyer it is.

I treated the listing like a form to fill out

The first version of my description was four sentences. It described what the skill was. It gave nobody a reason to believe me.

The version that sells today has three things the original didn't: a real sample transcript with the names changed, the exact output that transcript produces, and a short section literally titled "What this does NOT do." That last one felt counterintuitive — why advertise the limits? Because telling a skeptic where the edges are is the fastest way to make them trust everything else you've claimed. My "does not do" section says it won't handle transcripts over roughly 90 minutes well, and it won't magically fix garbage audio the transcription already mangled. Listing those two honest weaknesses did more for conversion than any adjective I could have added.

Write the listing for the most cynical person who might buy it. The fans don't need convincing.

I ignored my own changelog for two months

Here's the thing nobody mentioned to me: on a marketplace, the "updated" date is marketing. It's one of the few live signals a buyer has that there's a real human behind the listing who'll still be there when something breaks.

For my first two months I shipped nothing, because the skill worked and I'm lazy. Then I made a tiny change — I tightened the prompt so it stopped inventing action items nobody had actually assigned, a thing exactly four buyers had complained about. I wrote two sentences about it in the changelog and bumped the version. That week outsold the entire month before it. Same skill, slightly better, freshly dated. People weren't buying the improvement; they were buying the evidence that I hadn't abandoned them.

Now I ship something small every two or three weeks even when nothing's urgent. A clearer prompt. A better default. Two lines in the changelog. It's the cheapest marketing I have.

The one email that doubled everything

Around sale 200, I finally did the thing I'd been avoiding because it felt like homework: I emailed every single buyer. Three lines. What changed in the latest version, and one question — "what's the one thing you wish this also did?"

The replies became my roadmap, fully formed, for free. The most common request by a wide margin was "can it pull out the open questions — the stuff that got raised but never resolved?" I'd never have thought of it. I built it in an evening. It's now the feature people quote most in reviews, usually some version of "the open-questions list alone is worth the price."

The second most common request was a Slack-friendly output format, which I also shipped. Two features, both handed to me by the people already paying me, both from an email that took twenty minutes to write and send. I now send a version of that email after every meaningful update. The reply rate sits around 15%, which for a cold-ish email to busy people is absurd. They reply because almost nobody asks them anything.

The number that actually matters isn't 512

Everyone fixates on total sales. The number I watch now is refunds, and mine sits near 2%. That's the one that tells me whether the listing is honest. Early on, when the description oversold and the skill underdelivered, refunds spiked to almost 9% — every refund a small note that read "you lied to me a little." Tightening the listing to promise slightly less than the skill delivers dropped that number more than any feature ever did.

Low refunds compound, too. On a one-time-purchase marketplace there's no subscription to paper over a bad first impression — a buyer downloads the skill, runs it locally inside their own Claude, and either it earns its keep or they ask for their money back. Get that first run right and they leave the review that sells the next ten people. Get it wrong and you're refunding and burning the word of mouth in the same motion.

What I'd tell past me

• Launch at a price that embarrasses you slightly. You can always discount. Raising a price after you've anchored people at $7 is a much harder conversation.
• Write the listing for a skeptic, not a fan. Include what the skill won't do.
• Treat the updated date as a feature. Ship something small on a rhythm, even if it's just a clearer prompt.
• Watch refunds, not just sales. A 2% refund rate is worth more than a flashy launch week.
• Email your buyers and ask one real question. They'll hand you your next version for free, and they'll be flattered you asked.

The skill itself was never the hard part. I wrote it in an afternoon and barely touched the logic since. The work — the part I got wrong for months — was treating it like an actual product with an actual buyer on the other end, instead of a side effect I'd tossed over the wall and hoped would sell itself.