<div align="center">

<img src="assets/logo.png" alt="anti-sycophant" width="200">

# anti-sycophant

**No applause. Just the buyer you forgot to find.**

[](https://github.com/machinesoul11/anti-sycophant-ai-agent-skills/releases) [](https://github.com/machinesoul11/anti-sycophant-ai-agent-skills/stargazers) [](#quick-start) [](LICENSE)

*Skills that get AI assistants to stop agreeing with bad product ideas by default.*

</div>

Most chat LLMs default to enthusiasm. You say "I want to build an X that makes money," and the assistant gives you a stack pick, a name, and a starter schema — without ever asking whether anyone would buy it. That default is the single most expensive habit AI assistants have, because it converts an untested assumption into weeks of sunk work.

This pack ships three skills that interrupt that default. They work in Claude, Gemini, Codex/ChatGPT, Cursor, and any other tool that accepts a system prompt or rules file.

---

## See it work

With the pack installed, `/validate-idea "a note-taking app that makes money"` produces this instead of a stack recommendation:

> That's not an idea — it's a category plus a wish, and the wish is doing all the work. > > **The load-bearing assumption:** Someone will switch from a note-taking tool they already have, for free, to pay you for yours. Until you can name who and why, nothing else matters — not the stack, not the name, not the pricing page. > > **Why it's dangerous:** Note-taking is the most saturated consumer-software category that exists. You'd be competing with Ap