After a Boost from a Big AI Influencer, My GitHub Stars Finally Started Growing
When I wrote this title, my open-source project skill-base had just hit 95 GitHub stars.
For a top-tier project at a big company, 95 stars might be something you blink past. For an introverted indie dev who works a full-time day job and “casually” Vibe Codes side tools after hours, this rounds up to a highlight moment.
That unexpected bump got me thinking. Here are a few notes from the last stretch.
Vibe Coding lowers the bar—and flattens the moat
“Vibe Coding” comes from Andrej Karpathy. In short: you stop typing line by line. You describe what you want in a prompt and let AI (Claude Code, Codex, Cursor, and the rest) assemble the app for you.
The reality now: if you have an idea, almost anyone can ship something in days—or hours.
That sounds like a golden age for builders. The flip side is brutal: when everyone can build a spaceship overnight, space gets crowded fast. When the entry barrier disappears, the old technical moat goes with it.
Pain point 1: when everyone is shouting, who is listening?
That is the cold-start problem in plain terms.
When I finished the tool in March, I went all in: Product Hunt, Reddit, a proper WeChat article, even a live session with a tech media outlet. Product Hunt gave me one friendly upvote. Reddit barely rippled. GitHub stars stayed flat.
New accounts carry little weight. In the open ocean of the internet, the recommendation algorithms can treat your loudest shout like background noise. And plenty of people—developers especially—cannot bring themselves to post self-promo everywhere. After almost two months, the repo had only a dozen stars.
Then, not long ago, I nervously left a comment about skill-base under a WeChat post by Kazike (数字生命卡兹克), a well-known AI creator.
That brought in a lot of people with the same pain.
Stars jumped that day—from the teens to the seventies—and a long tail kept trickling in for a while after.
Pain point 2: surviving “Day 2” is where real products begin
Influencer traffic is a gust of wind: it passes. The AI era has no shortage of apps that flash bright and die.
Anyone can stitch together cool-looking demos with AI, ride information gaps or hype, and grab attention. Products that last usually face the Day 2 problem:
When novelty fades and people use the thing for real, does Vibe Coding–generated code start to crack?
What happens when a third-party API changes? Can your database hold up under load? And when AI-generated spaghetti grows faster than any human can read—how do you keep the project stable and maintainable?
Real product trust is not built from flashy UI. It is built by surviving time.
When users report bugs, will you show up? Do you have the time and patience to maintain and improve over the long haul? Every change risks subtle regressions—an old version breaks, data goes missing. That is where the gap opens up.
AI removed much of the grind of writing code and pushed engineering credibility and long-term patience to the front. Code is becoming a commodity. A developer’s judgment and reliability are the scarce goods.
Closing: stay dependable
AI makes cheap experiments easy: ship an MVP in hours or days, run a short test, and if it fails, pin hopes on the next idea. That rhythm will be normal for many people.
When non-technical creators declare that developers are obsolete, and titles like “I commanded 10 agents to …” or “Shocking: I used 100 agents to …” flood your feed, choosing long-term work and waiting for time to judge can feel luxurious.
Nobody scrolls past that without some anxiety. What matters, I think, is to stop feeding FOMO: build with your hands, polish deliberately, and prepare for the boring part—maintenance.
Vibe Coding something new feels great. Some people compare it to middle-aged fishing: quiet time alone, and you rarely come home empty-handed. Maintaining a product is the opposite— dull.
Getting something built your way is only step one of a long road—and that step keeps getting easier.
Without research, practice, and real conversations with users, sparks and fleeting ideas are hard to come by. (Large models mostly summarize the past and infer the future; they are not truly inventive on their own yet.)
I am grateful for those 95 stars on skill-base, and my head has settled down. (Early on I kept refreshing GitHub to see if the count moved.)
I do not want sensational headlines or a forced daily posting schedule. What I want is quieter: every bug fix, every restrained feature release, sending a bit of dependability to the folks following along—in China and abroad.
I am a frontend developer with a day job, Vibe Coding small tools in my spare time. If your team wants a low-friction way to distribute fragmented Agent Skills, check out skill-base on GitHub (95 stars at the time of writing). The skill-base site covers usage; for how to write Skills and roll them out on a team, the Agent Skills series on this site goes deeper. You can also find my other tools at reaidea.com—another one is coming soon.