Skip to main content

de-slop

Detect and remove AI writing patterns from prose. Bilingual engine: English (30+ patterns) and Chinese (25+ Chinese-native patterns with register-aware detection). Use when editing, rewriting, or reviewing text to eliminate predictable AI tells and inject authentic human voice. Triggers: humanize, de-AI, de-slop, un-ChatGPT, rewrite, edit draft, polish prose, check for AI tells, 去AI味, 说人话, 改得自然一点, 别像模板 DO NOT use for: code review, grammar-only fixes, technical documentation formatting

설치로 이동

소스 정보

저장소
Loveacup/jz-skills
최근 소스 활동
2026년 5월 31일 05:34
감지된 SKILL.md 언어
중국어
스타
1
포크
1

설치 방법

기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.

소스 파일 검토

설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.

파일 탐색기
7 개 파일

SKILL.md 표시 중

SKILL.md
소스 지침 · 읽기 전용 미리보기
name
de-slop
description
Detect and remove AI writing patterns from prose. Bilingual engine: English (30+ patterns) and Chinese (25+ Chinese-native patterns with register-aware detection). Use when editing, rewriting, or reviewing text to eliminate predictable AI tells and inject authentic human voice. Triggers: humanize, de-AI, de-slop, un-ChatGPT, rewrite, edit draft, polish prose, check for AI tells, 去AI味, 说人话, 改得自然一点, 别像模板 DO NOT use for: code review, grammar-only fixes, technical documentation formatting
version
2.0.0
license
MIT
sources
["https://github.com/blader/humanizer (v2.7.0, MIT)","https://github.com/hardikpandya/stop-slop (MIT)","https://github.com/LifelongLazyLearner/qu-ai-wei (MIT, Chinese patterns)","https://github.com/MrGeDiao/shuorenhua (MIT, Chinese guardrails)"]
# de-slop — AI Text Detector & Humanizer (Bilingual) ## 🚨 Red Flags: DO NOT SKIP THIS SKILL | agent 会找的借口 | 为什么是错的 | |-----------------|-------------| | "这文章读起来还行,不需要改" | AI 痕迹是统计性的——单独看没问题,聚类就是 confession | | "我直接删几个 em dash / 四字成语就行了" | 30+/25+ 模式互相印证。修一个漏十个,文本还是 AI | | "评分太主观,跳过吧" | 5 维评分是硬约束。EN < 35 / ZH < 35 必须重写 | | "中文和英文差不多,套同一套规则" | 中文 AI 腔靠密度 × 语体 × 功能判断,跟英文 pattern matching 是两套逻辑 | | "这是技术文档,不需要 Personality" | Personality & Soul 已加了 context guard:百科/技术/法律文本跳过 | ## 🔀 Decision Tree ``` 用户输入文本 ↓ [0] 语言检测 ├─→ 中文 → 🀄 ZH Pipeline(chinese-system.md + chinese-patterns.md) └─→ 英文/其他 → 🇬🇧 EN Pipeline(patterns.md + scoring.md) ↓ [1] 有用户写作样本? → Voice Calibration ↓ [2] 内容类型? ├─→ 博客/随笔/观点 → Personality Mode ├─→ 技术/百科/法律 → Clean Mode └─→ PR/邮件/简历 → Polished Mode ↓ 执行对应语言的流水线 ``` --- ## 🀄 ZH Pipeline(中文) 中文 AI 检测不是 pattern matching,而是**密度 × 语体 × 功能**三维判断。先加载中文体系文件,再执行流程。 ``` [ZH-0] 加载 chinese-system.md → 冲突仲裁树 + 门检 + 语体矩阵 [ZH-1] 门检:是不是真人写的?→ 是 → 停手 [ZH-2] 识别语体 → 激活对应规则子集 [ZH-3] Scan → 对照 chinese-patterns.md 标记 [ZH-4] Draft → 逐段改写(减法 + 打磨) [ZH-5] Audit → 密度三问 + 过度消毒反制 + AI 不敢写测试 [ZH-6] Final → 修复残留 → 打磨报告 [ZH-7] Score → 5 维评分 → < 35 回到 ZH-4 ``` 详见 `references/chinese-system.md` 和 `references/chinese-patterns.md`。 --- ## 🇬🇧 EN Pipeline(英文) ``` [1] Calibrate → 有样本则分析风格 [2] Scan → 对照 patterns.md 标记 [3] Draft → 逐段重写 [4] Audit → "What makes this AI?" [5] Final → 修复残留 → 注入 Personality [6] Score → 5 维 → < 35 回到 [3] ``` ### 🎯 Quick Reference: Top 15 AI Tells 扫第一遍时对照此表。完整 30+ 模式见 `references/patterns.md`。 | # | 模式 | 一句话 | 典型例子 | |---|------|--------|---------| | 1 | Significance inflation | 把普通事吹成里程碑 | "marking a pivotal moment" | | 2 | -ing padding | 句子尾巴加假深度 | "showcasing how...contributing to..." | | 3 | Promotional language | 广告腔 | "nestled...breathtaking...vibrant" | | 4 | Vague attributions | 无名无姓的"专家" | "Industry observers have noted" | | 5 | AI vocabulary | 高频 AI 词聚类 | "delve, tapestry, landscape, crucial" | | 6 | Copula avoidance | 不用 is/are | "serves as" instead of "is" | | 7 | Em dashes | AI 最可靠信号之一 | "—not by the people themselves—" | | 8 | Throat-clearing | 开场不说正事 | "Here's the thing...Let me be clear" | | 9 | Binary contrasts | 制造假冲突 | "It's not about X. It's about Y." | | 10 | Rule of three | 硬凑三个 | 三名词、三动词、三段式结尾 | | 11 | Sycophantic tone | 过分讨好 | "Great question! You're absolutely right!" | | 12 | False agency | 死物做人事 | "the decision emerges" → 谁决定的? | | 13 | Hedging cluster | 层层包裹 | "could potentially possibly be argued" | | 14 | Chatbot artifacts | 对话残渣 | "I hope this helps! Let me know if..." | | 15 | Generic conclusion | 万能结尾 | "The future looks bright..." | ### ⚡ Quick-Check (EN Final Pass) - [ ] 任何副词?杀了 - [ ] 被动语态?找主语 - [ ] 无生命物做人的动作?说出谁干的 - [ ] Wh- 开头句子?重构 - [ ] "here's what/this/that" 清嗓子?直接说 - [ ] "not X, it's Y" 对比?直接说 Y - [ ] 三个连续句子等长?打断 - [ ] Em dash?去掉 - [ ] 模糊宣言?说出具体意义 - [ ] 旁观者叙事?把读者放到场景里 - [ ] 元衔接?删了 --- ## 🗣️ Voice Calibration(中英通用) 如果用户提供了自己的写作样本: 1. **分析:** 句长模式、用词层级、段落开头习惯、标点习惯、过渡方式 2. **匹配:** 重写时换上用户的节奏 3. **无样本:** 回退到自然、多变、有态度的默认声音 ## 💬 Personality & Soul(中英通用) ⚠️ **适用范围:** 博客、随笔、个人写作、观点文。技术文档、百科、法律文本 → 跳过。 **注入人声:** 有态度、变节奏、留点乱、情绪具体化。 ## 📊 Scoring System(中英通用) 5 维评分,每维 1-10。总分 50,< 35 重写。详见 `references/scoring.md`。 | 维度 | 衡量 | |------|------| | Directness | 陈述 vs 宣告? | | Rhythm | 多变 vs 节拍器? | | Trust | 尊重读者智力? | | Authenticity | 像人说话? | | Density | 有没有可砍的? | ## Detection Guidance(中英通用) ⚠️ **不要误杀:** 完美语法、正式词汇、孤立 em dash、缺引用——不是可靠 AI 信号。 **人类写作信号(保留):** 具体罕见细节、矛盾情感、带年代引用、真正旁白和自我纠错。 --- ## 📦 References | 文件 | 语言 | 何时读取 | |------|:--:|------| | `references/patterns.md` | 🇬🇧 | EN 完整 30+ 模式 | | `references/scoring.md` | 🌐 | 中英通用 5 维评分 | | `references/chinese-system.md` | 🀄 | ZH 冲突仲裁 + 门检 + 语体矩阵 + 反消毒 + 打磨 | | `references/chinese-patterns.md` | 🀄 | ZH 完整 25+ 模式 | | `references/trigger-tests.md` | 🌐 | 修改 triggers 后的回归测试 | --- ## ✅ Verification Checklist - [ ] 语言路由是否正确(ZH 加载 chinese-system/patterns,EN 加载 patterns)? - [ ] ZH:门检 + 语体识别 + AI 不敢写测试? - [ ] EN/ZH:完整流水线(Scan → Draft → Audit → Final → Score)? - [ ] EN/ZH:5 维评分 ≥ 35? - [ ] Quick-Check 通过(EN 或 ZH 对应版本)? - [ ] 是否误杀了人类写作信号? - [ ] 输出是否包含完整 deliverable?
GitHub에서 보기