no-ai-slop

---
name: no-ai-slop
description: "Use when a draft needs to sound human — edits out 20+ named AI-slop patterns (colon reveals, binary contrasts, banned words) while preserving the writer's voice, or detects and names the patterns without rewriting."
source: https://github.com/petergyang/no-ai-slop
generated: 2026-07-23T01:35:11.035Z
category: concept
audience: creators
---

## When to use

- A draft sounds AI-generated and should read like a human wrote it, without losing the writer's voice
- Editing a blog post, launch note, or thread to be clearer, more direct, and more opinionated
- Auditing a piece for AI-slop patterns in detect mode — name and quote each pattern, no rewrite
- Deciding whether tics like 'delve', colon reveals, or 'not X, it's Y' contrasts should be cut

## Key concepts

### Two jobs: Edit vs Detect

Edit is the default: make the minimum effective edit and return the draft plus a 'What changed' section. Detect answers 'is this AI slop?': name each pattern from the skill that appears, quote the line, give the fix in a few words — no rewriting, no scoring, no guessing whether AI wrote it.

### Minimum effective edit

Fix AI patterns, errors, repetition, and unclear passages; leave strong human sentences alone. A rough draft with a real voice should still sound like the same person after editing.

### Voice signals

Before editing, identify the core point and 3–5 traits to preserve — vocabulary, cadence, bluntness, humor, uncertainty, digressions. The note stays internal; distinctive lines are never rewritten merely for consistency.

### Banned words

Cut outright: delve, foster, leverage, utilize, facilitate, empower, streamline, robust, cutting-edge, paradigm shift, game changer, tapestry, realm, transformative, elevate, embark, supercharge, harness, ever-evolving. Often-empty adverbs and phrases ('just', 'it's worth noting') are cut only when they add nothing.

### Pattern catalog

20+ named patterns with before/after fixes: binary contrasts, throat-clearing openers, faux-insight setups, colon reveals, importance puffery, weasel attribution, synonym cycling, negative listing, dramatic fragmentation, rhetorical setups, fake-profound kickers, summary-recap endings, formatting slop, em-dash clusters.

### eval.md self-check

After editing, the draft is checked against the repo's eval.md; any failing check sends the editor back to fix and re-check before output. The workflow ends with the full edited draft plus a short 'What changed' section.

## Gotchas

- Detect mode never rewrites, scores, or guesses whether AI wrote a piece — detectors guess; named, quoted patterns are evidence the user can check
- Keep 'I think', 'maybe', 'honestly' when they carry real uncertainty or the writer's spoken rhythm — only cut them when empty
- Don't rewrite a fake-profound kicker into a better metaphor — delete it and end on the clearest concrete sentence already in the draft
- Em dashes: none in short copy; 1–2 in longer drafts only when they clearly beat commas, periods, or parentheses
- Never invent claims, examples, stats, or sources — name the source or ask the user instead of smoothing over the gap
- Front-load conclusions only when it improves clarity; don't force every paragraph into the same point-detail-background shape
- If audience or format is unclear, ask exactly one question before editing: who is this for and where will it be published?

---
Generated by SkillMake from https://github.com/petergyang/no-ai-slop on 2026-07-23T01:35:11.035Z.
Verify against source before relying on details.

File: ~/.claude/skills/no-ai-slop/SKILL.md