Slop Cop

A browser-based editor that detects stylistic traces and structural patterns left by LLMs in real time.

The core is a two-layer detection approach.

• Client-side instant detection: 35 rules implemented with regex and structural analysis run on every keystroke with about 350ms debounce.
• Optional semantic analysis: provides two manual analyses using the Anthropic API.
  • Fast pass: uses Claude Haiku to catch sentence- and paragraph-level patterns.
  • Deep pass: uses Claude Sonnet to look at patterns that emerge across the whole document, such as repetitiveness, dilution of a single point, and repeated summarization.

Detection targets include exaggerated modifiers, unnecessary substitution with elevated vocabulary, filler expressions at the start of sentences, excessive contrast constructions, list-style development, and unnatural concluding phrases. For example, these are classified under rules like Overused Intensifier, Elevated Register, Connector Addiction, and False Conclusion.

How it works is simple.

• There is no separate server.
• API calls go directly from the browser to Anthropic.
• The API key is stored only in localStorage.
• Large documents are automatically split into chunks and processed in parallel.

The goal is not to guess "whether this text was written by AI," but to surface whether it reads like AI right at the editing stage, immediately. Users can look at the highlights and instantly decide which expressions to keep and which to remove.