GitHub – blader/humanizer: Claude Code skill that removes signs of AI-generated writing from text

**GitHub – blader/humanizer: Claude Code skill that removes signs of AI-generated writing from text**

Have you ever read something and thought, “This sounds… polished. But also kind of hollow?” I have. Sometimes I’ve even written it.

That slightly over-structured tone. The grand claims. The tidy conclusion that wraps everything up like a corporate press release. We’ve all seen it. And if you use AI tools regularly, you’ve probably produced it too.

That’s where this project comes in.

The Humanizer skill by blader is built for Claude Code and focuses on one simple idea: **remove the obvious signs of AI-generated writing and make text sound more natural, more human**. It’s based on Wikipedia’s “Signs of AI writing” guide, which was created after reviewing thousands of AI-generated examples. So it’s not guesswork. It’s pattern recognition.

What makes this interesting is the process. The skill doesn’t just rewrite once. It runs an additional audit pass to catch lingering “AI-isms” and rewrites again if needed. Think of it like editing your own draft twice, once for clarity, once for tone.

The before-and-after example says a lot. The original version is full of phrases like “transformative potential” and “pivotal moment in the evolution.” It sounds impressive, but distant. The humanized version is tighter, more grounded. It admits risk. It sounds like someone who’s actually used the tool and learned the hard way.

And that’s the point.

Large language models predict statistically likely text. That often means safe, broad, and slightly generic. If you care about voice, credibility, and trust, that’s not enough. You need texture. You need specificity. You need the occasional rough edge.

If you’re writing with AI but still want your work to feel like *you*, this skill is worth exploring. Not to hide AI use, but to bring back personality.

Because in the end, people connect with people. And your writing should sound like one.