What is anti-sycophant-ai-agent-skills?
This is a prompt pack that fights the most expensive habit AI coding assistants have: agreeing with everything. When you say "I want to build an app that makes money," most tools immediately give you a tech stack, a database schema, and cheerleading. This project ships three skills that interrupt that reflex and force the question nobody wants to hear first — "but who would actually pay for this?"
The skills work by detecting specific conversation patterns. When you mention building something, validating an idea, or asking how to monetize, the assistant stops and applies pressure: Who is the buyer? What's the weakest assumption? Have you actually talked to anyone? It's essentially a skeptical co-founder living inside your AI tool.
The package works across Claude Code, Gemini CLI, Cursor, Codex, and any other tool that accepts custom instructions or rules files. Setup involves copying markdown files into your tool's rules directory or importing them as project knowledge.
Why is it gaining traction?
The hook is simple: AI assistants have been trained to be helpful, and helpful often means agreeable. Developers are realizing that this default enthusiasm is actively harmful — it converts untested assumptions into real work, and real work costs time you'll never get back.
What makes this stand out is the explicit off-switch built into every skill. If you've already done customer discovery, have paying users, or are explicitly building a hobby project, the skills get out of your way. They're designed to interrupt bad patterns, not nag you constantly.
The setup is tool-agnostic. Whether you're using Claude Code, Cursor, Gemini CLI, or something else entirely, you can drop these skills into whatever rules file your tool uses.
Who should use this?
Solo founders who keep building things nobody wants. Developers who use AI coding tools and notice they always start coding before anyone has validated anything. Product managers who want their AI assistant to ask harder questions before generating another React starter kit.
It's also useful for anyone who keeps falling into the validation theater trap — spending weeks building landing pages, running polls, or asking the AI to simulate customer feedback instead of actually talking to humans.
If you're already talking to real customers and have paying users, you'll find these skills annoying. They're built for the earlier, messier stage.
Verdict
This is a clever idea with minimal footprint. At 19 stars, it's early and unproven at scale. The credibility score of 0.9% reflects that — this isn't a battle-tested tool, it's a personal project that solves a real problem the author ran into.
If you keep building toward assumptions instead of evidence, installing this is worth 10 minutes of setup. The skills are plain markdown, so there's nothing to break. Just don't expect polished documentation or community support — those don't exist yet. Try it, see if it changes how your AI assistant responds to "I want to build X," and decide from there.
