unslopify
Audit and rewrite text to remove AI-writing patterns. A deterministic Python core names the failures. An agent skill runs the rewrite loop until a fresh-context judge cannot tell a model touched the text.
Before
In today's fast-paced digital world, this release is not just an update. It is a testament to our team's unwavering commitment, ensuring a seamless experience for every user.
After
This release fixes the checkout crash and cuts page load from 3 s to 1 s. Both changes came out of the June incident review.
$ unslopify draft.md
draft.md:1: [scene-setting] "In today's fast-paced digital world"
draft.md:1: [inflated-contrast] "is not just"
FAIL draft.md: 2 findingsWhat it does
Generated text has tells: manufactured contrast, fake authority, tacked-on benefits, vocabulary no person reaches for. unslopify names 27 of them across four categories, finds each one with line and span, and produces rewrite instructions. A phrase bank remembers the wording of everything you have shipped, so no two documents lean on the same stock phrasing.
Built for agents
The repo ships a SKILL.md that plugs into Claude Code, Cursor, or any harness with skills. It runs the full loop: audit, plain-language rewrite with every edit logged as a typed event, uniqueness check, then a judge in a separate context that only sees the finished text. The audit is pure Python with no model calls, so the same input always gives the same report.