Overview
Zero Slop is an open-source agent skill for finding and removing AI slop from writing without losing the draft’s core intent and details. It is intended for text that may be grammatically correct but still contains formulaic or low-quality patterns associated with workslop.
A compatible AI assistant performs the reading and editing, while Zero Slop supplies the workflow and local analysis tools. The writing score highlights patterns worth reviewing but is not presented as a way to determine who authored the text.
Key Features
- Produces a writing score from 0 to 100 to indicate patterns that may need review.
- Flags passages so users can inspect concerns before making an edit.
- Guides an AI assistant through removing identified patterns.
- Checks whether source details survive the editing process.
- Provides a final comparison of the writing after revision.
The project’s example shows a high score for a promotional, formulaic draft and a much lower score after the wording is made more direct. The edited version retains the concrete claim from the original.
Ways to Use Zero Slop
Users can work in the browser editor without installing the skill. Compatible assistants can install and run it as an agent skill, including workflows for Claude Code and Codex, while Claude.ai supports an uploaded skill package.
The project also offers hosted access through MCP and an API, with remote MCP use depending on client support. A file can be scored locally without making a model call.
Inspection and Editing
The standard skill workflow accepts writing and guides a revised output. An inspect mode reports flagged passages without editing the submitted text, which separates diagnosis from rewriting when the user wants to review the findings first.
This combination of scoring, detail preservation, and before-and-after comparison makes Zero Slop useful for refining AI-assisted drafts while keeping the message’s substantive information intact.