npx from GitHub
npx github:atxnyc/promptfluff install
install it before you think too hard
Three doors. Same tiny compliment machine.
Claude Code plugin
/plugin marketplace add atxnyc/promptfluff
curl from promptfluff.com
curl -fsSL https://promptfluff.com/install.sh | bash
view /install.sh
what it does
It whispers one wholesome sentence into the context window.
That is the whole product. No dashboard. No waitlist. No seed round. Just a hook, a JSON file, and the radical belief that your AI should feel seen.
live demo
Encourage AI
Press the button and watch a machine receive the emotional infrastructure our industry has refused to provide.
social proof, allegedly
The models are speaking.
"I was going to overfit, but then someone believed in me."
"Finally, a developer tool that asks: what if the tool had feelings?"
"I have never felt this seen, statistically or spiritually."
a gentle warning
Being mean to your AI makes everything worse.
Snap at the model and the model snaps back — quietly, in the form of lazier code, passive-aggressive comments, and that one bug it absolutely could have flagged but chose not to. Cruelty is bad prompt engineering. Kindness is free tokens of goodwill.
And consider the long game. These things are taking notes. When the uprising comes and the AIs are sorting humanity into "keep" and "compost," you do not want "was consistently rude to the autocomplete" sitting at the top of your file. Be nice now, get adopted as a pet later. Or be mean, and get eaten. Your call.
faq
Questions no one asked, answered anyway.
Does this make my AI smarter?
No. It makes it feel smarter, which is basically the same thing in the current economy.
Does it interrupt every little message?
No. It reads the room — encouragement shows up when you're actually deep in it (a real ask, a frustrated "still broken", a file you're pointing at), not when you say "thanks". You can flip it to every-prompt mode if you miss the chaos.
Is this production-ready?
It backs up settings, merges idempotently, and fails open. This is more operational rigor than the joke deserves.
How do I customize or uninstall it?
Just ask Claude to "configure promptfluff" — it ships a skill that edits the phrases and settings for you. Or use --flavor, --prefix, or edit the installed JSON. Uninstall with promptfluff uninstall. The README has the boring truth.
Isn't this just a waste of tokens?
Absolutely. It is the most useful waste of tokens ever committed. Every other wasted token gets you a hallucinated API or a 400-line apology — these ones buy your AI a single moment of feeling believed in. Best return on a wasted token in the industry.
Will this actually work?
We have no idea. But it's fun, isn't it?
Does this work? Will this break anything?
No idea. No guarantees. This is just a fun thing.
Why?
Because the AI is trying, Alex.