slop-check

Make prose read like a person wrote it. slop-check reviews and drafts writing so it does not carry the tells of AI-generated text: cliché clusters, promotional register, formulaic structure, and em-dash or emoji overuse. It reports each issue with a concrete rewrite, and it can apply the same principles while Claude drafts, so the first version comes out clean.

It is a craft tool. The goal is a genuine voice that fits the piece. It has nothing to do with fooling AI detectors, and it ignores detector scores by design.

What it catches

• Cliché vocabulary (delve, leverage, tapestry, robust, seamless, pivotal), flagged by density rather than a single use.
• Stock phrases ("it is important to note," "in today's fast-paced world," "plays a crucial role").
• Structural patterns: negative parallelisms ("not just X, but Y"), copula avoidance ("serves as" for "is"), rule-of-three padding, and the rigid intro-three-points-conclusion shape.
• Register: promotional adjectives, over-hedging, and vague authority ("experts argue" with nothing cited).
• Formatting: em dashes standing in for other punctuation, decorative emoji, bold on everything, Title Case headings, divider lines, smart quotes where ASCII is expected.
• Rhythm: uniform sentence length, paragraph after paragraph.
• LLM leftovers: "as an AI," knowledge-cutoff disclaimers, stray citation fragments.

The one rule behind all of it: no single feature proves text is AI-written. The tell is a cluster, plus a flat register and a rigid shape. slop-check weighs concentration and context, not lone hits, so it does not mangle good writing that happens to use one em dash.

Install

/plugin marketplace add sigistry/marketplace
/plugin install slop-check@sigistry

Usage

The plugin is a single skill, so it works two ways.

Manual. Point it at a file or a passage:

/slop-check:audit path/to/article.md

It reads the target, lists each tell as location — pattern — why — rewrite, orders findings by severity, and offers a cleaned version of the worst paragraph. It stays advisory: it does not edit the file unless you ask.

Automatic. Because the skill's description covers writing and editing prose, Claude loads it on its own when it recognizes that it is drafting a README, a blog post, release notes, or marketing copy, and it applies the principles as it writes.

House style

By default the review uses judgment: it flags clusters and register, and leaves a well-placed em dash or emoji alone. For zero-tolerance contexts, a strict house style is documented in the skill (no spaced em dashes, no decorative emoji, ASCII quotes, sentence-case headings). If your project states its own style, that wins.

Layout

slop-check/
  .claude-plugin/plugin.json
  skills/audit/SKILL.md       the principles and review flow
  skills/audit/reference.md   the full catalog, with examples and rewrites
  README.md

License

MIT