Unslop — AI skill for Claude Code

The problem

Most AI-generated social content sounds the same. Sentences run the same length. "Unlock," "revolutionize," and "testament to" show up whether or not they mean anything. The hook doesn't hook. Nothing in it is specific enough to actually be about anything. That's not an accident — it reads like it was written for no one in particular, because it was.

Unslop is a [Claude Skill](https://github.com/anthropics/skills) that takes that draft, works out exactly which patterns are making it generic, and rebuilds it around a real hook-hold-payoff structure tuned to how the target platform actually keeps or loses attention.

What makes this different

This isn't an AI-detector-evasion tool. Detectors change constantly and have high false-positive rates on non-native English and technical writing, so building around beating them is a losing bet, and a slightly dishonest one.

Unslop works from the psychology instead: curiosity gaps (Loewenstein), the Zeigarnik effect, emotional arousal and sharing (STEPPS), and hard platform-retention data — TikTok's first-second attention gate, LinkedIn's 85% muted-viewing rate, Instagram's "caption valley." It also won't invent a stat, credential, or personal story to make a hook land harder. See [Guardrails](./SKILL.md#guardrails).

Install

**Claude Code (recommended)**

git clon