ai-slop-killerlisted
Install: claude install-skill mrhinkle/aienterprise-skills
# AI Slop Killer
The last skill to touch prose before it ships. Whatever drafted or revised it, this runs the passage through one job: make it read like a person wrote it on purpose, then prove what it removed. Works on any genre — business writing, marketing, docs, essays, fiction.
The pattern catalog is built from documented sources (Wikipedia's *Signs of AI writing*, the tropes.fyi directory, working editors' guides — see `reference/slop-catalog.md`), not one person's taste. It ships with sensible defaults and an optional per-user config for house rules.
## The loop: Detect → Score → Fix → Prove
1. **Read the passage cold**, start to finish, before touching anything. Slop clusters — you're looking for stacked tells, not one stray word.
2. **Detect.** Run every check in `reference/slop-catalog.md`. Match on the *pattern*, not exact wording. A single instance of most tells is human; the tell is the *repetition and clustering*.
3. **Score.** Compute the slop score using `reference/scoring.md`. If it's ≤5 per 500 words, say "clean" in one line and stop — don't manufacture edits.
4. **Fix in place.** Rewrite the tells. Preserve the author's meaning and voice. If a voice/style skill or a config is available (see below), honor it. Never flatten a correct term of art into vagueness.
5. **Prove.** Return the cleaned passage, then the before→after score and an itemized list of what was cut and why — plus anything left in on purpose.
## The non-negotiables
- **Kill negative par