humanizer vs stop-slop
Portable agent skill that strips the signs of AI-generated writing — 25 pattern categories from Wikipedia's 'Signs of AI writing' guide, with a two-pass self-critique loop and voice calibration. — versus — A skill file that removes AI tells from prose: ~30 banned phrases across 7 categories, 8 banned structural patterns, and a 50-point scoring rubric that forces revision below 35. Zero dependencies.
The two anti-slop skills: stop-slop enforces hard bans and a scoring rubric; humanizer runs a two-pass self-critique loop and voice calibration. Rules vs reflection — same slop, different weapon.
| humanizer | stop-slop | |
|---|---|---|
| Stars | 34k | 15k |
| Forks | 3.1k | 1.1k |
| Language | Python | — |
| License | MIT | MIT |
| Last activity | 18 days ago | 4 months ago |
| Topics | skills | skills |
| Curated connections | 2 | 1 |
humanizer — the curator's take
The two-pass loop is why it beats phrase blocklists: after rewriting, it asks 'what still reads as AI?' about its own output and revises again — and it pushes you to ADD voice, not just delete tells, which is the half most anti-slop tools skip. Plain SKILL.md, runs in any harness. When NOT: it edits prose, not facts — hallucinations survive humanization untouched; and voice calibration needs your writing samples or you get generic-human instead of you.
stop-slop — the curator's take
The sharpest ideas in the anti-slop space live here: 'false agency' detection (naming the human instead of 'the data tells us') and a scoring rubric that makes revision mechanical rather than vibes. Em dashes categorically banned — it has opinions. When NOT: the hard bans ARE the product — if your voice legitimately uses em dashes or fragments, this fights you; it's rules-only with no self-critique loop, so it catches listed patterns and misses novel ones.