LLM Slop Detector 作者： Mathias. A

LLM Slop Detector highlights the things that give AI-generated writing away: invisible Unicode smuggled inside text, em dashes and curly quotes that spell-checkers ignore, and phrases like "delve into", "tapestry of", "it's worth noting", or "game-changing" that appear far more often in LLM output than in human writing.

It works in two ways:

As you write
- Scans every textarea, text input, and contenteditable on any webpage.
- Inline wavy underlines mark flagged words and phrases. A floating badge shows the count. Click it for per-finding explanations.
- One-click fixes for deterministic characters (em dash -> hyphen, curly quotes -> straight, zero-width spaces -> deleted). The fix goes through the host editor's native undo stack, so Cmd+Z works.
- Tested in Gmail compose, Proton Mail compose, GitHub issue forms, Reddit, Substack, and anything else built on plain contenteditable.

As you read
- Click the "Scan this page" button in the toolbar popup to highlight slop in any article, blog post, or thread you're reading. Wavy underlines go under flagged words; a floating results panel lists every finding and jumps you to it on click.
- One-shot, click-to-scan only. No background scanning.

Rules are configurable
- ~40 built-in core rules plus eleven opt-in packs covering academic writing, general LLM cliches, fiction tells, Claude-specific mannerisms, structural patterns ("not X but Y"), invisible-Unicode security threats, and model-specific tells for Gemini, DeepSeek, Llama, Qwen, and Grok -- 500+ curated regex patterns in total.
- Enable only the packs you care about. Per-site disable in one click.

Privacy
- Nothing you type, read, or scan is ever transmitted, logged, or persisted. All scanning runs locally inside the extension's sandboxed JS.
- No telemetry, no analytics, no remote rule updates. The rules are bundled at build time and ship with each extension release.
- Storage is chrome.storage.local only; your settings don't leave this device.
- No "cloud assist" toggle. Never has been. Never will be.

Open source (MIT) at github.com/mandakan/llm-slop-detector. Also available as a VS Code extension, an npm CLI, and a web playground for paste-and-check. Same rule engine across all of them.