360 Secure SkillHub
360 安全技能中心

Anti AI Slop Writing Gate

该 Skill 提供编辑、检测和发送前三种模式,用于识别或最小化处理公式化 AI 文风,去除空话、企业套话和空洞铺垫,同时保留作者观点、词汇、节奏与事实依据;检测模式只列出模式、引用原句并给出简短修改建议,不重写、不判断作者身份或 AI 生成概率。

mwclaw
mwclaw
热度 15.00
最近更新 2026-08-20 01:21:37
分类:效率依赖:读写文件应用场景:校对润色 / 文案创作

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add anti-ai-slop-writing-gate
Metadata
name
anti-ai-slop-writing-gate
description
Detect or minimally edit formulaic AI-style prose while preserving the writer's voice.
author
Mike Winkler

Anti AI Slop Writing Gate

Use this skill to edit a draft, identify checkable AI-slop patterns without rewriting it, or run a silent pre-send gate on substantive prose.

Do not use stylistic patterns to claim that AI wrote a text. The skill evaluates the writing, not its authorship.

Modes

  • Edit: Make the minimum effective edit. Preserve the writer's point, vocabulary, cadence, humor, uncertainty, edge, and useful roughness.
  • Detect: Name the patterns present, quote the exact line, and give a short fix. Do not rewrite or assign an AI-likelihood score.
  • Pre-send: Silently inspect a substantive reply before delivery and fix supported failures. Do not mention the audit unless asked.

Use pre-send mode for interpretations, advice, research synthesis, consequential explanations, and other developed replies. Skip it for acknowledgments, exact factual one-liners, raw tool output, or urgent safety instructions.

Workflow

  1. Read the full draft. Identify the audience, job, core point, intended action, and existing voice signals.
  2. Load references/patterns.md.
  3. If the audience, format, or reader action is materially unclear, ask one focused question. Otherwise proceed.
  4. In detect mode, report the evidence and stop.
  5. In edit or pre-send mode:
    • remove filler, fake profundity, corporate abstraction, unsupported puffery, and repetitive cadence;
    • state the useful claim directly instead of staging it with throat-clearing or an unnecessary contrast;
    • preserve strong human sentences, specific facts, and useful roughness;
    • add concreteness, ownership, constraint, mechanism, or tradeoff only when supported by the draft;
    • never invent or silently strengthen claims, examples, numbers, quotations, sources, opinions, or emotion;
    • do not over-compress merely to eliminate a surface pattern.
  6. Check the result against references/evals.md. Fix every failure before returning or sending it.

Output

Edit

  • Clean Version
  • What Changed — short and specific
  • Remaining Risk only when meaning, sourcing, or factual support still needs the writer

Detect

For each finding:

  • pattern name
  • exact quoted line
  • short fix

Offer to edit after the audit.

Pre-send

Return only the corrected substantive reply.

Standard

Passes when the result is clear, specific, natural aloud, and recognizably the writer's.

Fails when it merely swaps buzzwords, erases personality, invents support, turns the piece into generic professional prose, or lets a known pattern through because the skill was treated as passive guidance.