Lieflat Less AI Tone

三层去 AI 味：脚本检测出分级报告（A 机械改/B 建议改/C 只报告/X 已否决/F 指纹）+ 语义层判读（同义重命名/衔接连环/结论冒号/同构三连），先向用户交付专业、可读的改前报告，列出本篇指标、研究显著性、原文例句和具体改法；收到用户回复后按白名单规则改写，复测并生成前后对比报告（默认 Markdown，与文稿同格式）。规则基于两轮实测（629 篇对照 + 270 对同题同事实配对复测）整理。适用于写作完成后的成稿清理。

4,684 stars 265 forksAI, LLMs & Agents

Install

skills CLI (Claude Code, Cursor, Codex and more)

npx skills add larashero3-dotcom/lieflat-less-ai-tone --skill lieflat-less-ai-tone

Manual install

git clone --depth 1 https://github.com/larashero3-dotcom/lieflat-less-ai-tone.git /tmp/lieflat-less-ai-tone
mkdir -p ~/.claude/skills
cp -r /tmp/lieflat-less-ai-tone/. ~/.claude/skills/lieflat-less-ai-tone

Personal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.

From the SKILL.md

# 去AI味 v2

识别并改写文本中的 AI 写作痕迹。工作顺序是**先检测、用指标与例句讲清发现和改法、等待用户回复、再改写与复测交付**。报告要专业且容易读懂：说明哪些特征在研究语料中显著，本篇实际命中了多少，以及为什么这些位置值得改。模型负责逐条筛选，不把几十个候选和比喻清单交给用户逐项决策。用户回复「按建议改」即可继续，也可以一句话调整范围。规则基于两轮实测：第一轮 629 篇对照语料，第二轮 270 对同题同事实配对复测 + 43 万次语义判断。配对复测推翻了 v1 的三条规则（见「X 级清单」），这部分规则只出现在检测报告里，永不执行。

检测分三层分工：**数得出符号的归脚本**（detect.py 正则/结构指标），**要理解句意的归语义层**（本会话模型按 [references/semantic-criteria.md](./references/semantic-criteria.md) 的封闭判据逐条判断），**机器间共识不够的归人审**（比喻贴切度、暗喻、拟人化喻体——报告里只列人审区，模型不判）。

## 工作流（必须按顺序）

### 第一步：脚本检测

```bash
python3 scripts/detect.py <文件或目录> [--max-hits 5] [--json]
```

拿到分级报告。各级含义和判级依据见 [DETECT.md](./DETECT.md)。**改前这次务必带 `--json` 存档**（第四步对比报告要用）。

### 第 1.5 步：语义层判读（单篇必跑，目录批量可选）

```bash
python3 scripts/detect.py <文件> --candidates --out /tmp/candidates.json
```

对抽出的四类候选（同义重命名段 / 段间衔接接口 / 冒号句 / 三连同构组），按 [references/semantic-criteria.md](./references/semantic-criteria.md) 的判据逐条给出 `label`/`confidence`/`reason`，写回 JSON。判断纪律（43 万次打标的教训）：只按判据判不按整体感觉、每条独立判断不看全文、拿不准选兜底标签、禁止自造标签。

单篇 3000 字约 40-80 条候选，会话内直接判，token 开销可忽略。判完按判据文件末尾的汇总口径算四个数：重命名 次/段、最长连环、结论冒号占比、高度同构三连数。

### 第二步：给用户看检测结果和建议，等待回复

读完脚本报告和语义层四项汇总后，向用户交付一份**专业、可读的改前检测报告**，并在此结束本轮。篇幅随文章和问题数量而定；不能只给一句感想，也不能把原始脚本输出原样贴给用户。报告至少包含：

1. **方法和口径**：说明脚本统计了什么、语义层判读了什么，给出本篇字数、段落数和候选数量。明确区分“研究语料中的统计显著”和“本篇单次命中”；没有做单篇显著性检验就不能宣称本文统计显著，也不能据此断言作者是 AI。
2. **指标结果与命中去向**：用一张紧凑表列出本篇命中的第一层跨模型指标和第二层单模型指纹。能定位到句子的项目写明**原始命中数、有效命中数、拟改数、保留数**，以及本篇密度、研究参照和证据强度；句长 CV、结构方差等整篇分布指标只报实测值与处置，不强造逐句命中数。原始命中为零也可简列。即使未达处理门槛或决定保留，也不得从报告中删掉这一项；不同轮次计数口径不同，不得混算倍数。

Read the full skill on GitHub

Details

Skill name: lieflat-less-ai-tone
Repository: larashero3-dotcom/lieflat-less-ai-tone
Repo updated: 27 October 2026

FAQ

What does the Lieflat Less AI Tone skill do?

How do I install the Lieflat Less AI Tone skill?

skills CLI (Claude Code, Cursor, Codex and more): npx skills add larashero3-dotcom/lieflat-less-ai-tone --skill lieflat-less-ai-tone. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.

Is Lieflat Less AI Tone popular?

It lives in larashero3-dotcom/lieflat-less-ai-tone, which has 4,684 GitHub stars and 265 forks.

Related skills

LLM Council

2.9k

aiwithremy/claude-skills-llm-council

Run any question, idea, or decision through a council of 5 AI advisors who independently analyze it, peer-review each other anonymously, and synthesize a final verdict. Based on Karpathy's LLM Council methodology. MANDATORY TRIGGERS: 'council this', 'run the council', 'war room this', 'pressure-test this', 'stress-test this', 'debate this'. STRONG TRIGGERS (use when combined with a real decision or tradeoff): 'should I X or Y', 'which option', 'what would you do', 'is this the right move', 'validate this', 'get multiple perspectives', 'I can't decide', 'I'm torn between'. Do NOT trigger on sim

AI, LLMs & Agents248

Agents SDK

3.2k

cloudflare/skills

Build, debug, or review Cloudflare Agents SDK applications using the agents package.

AI, LLMs & AgentsIn 1 list336

Karpathy LLM Wiki

2.6k

Astro-Han/karpathy-llm-wiki

Use when building or maintaining a personal LLM-powered knowledge base. Triggers: ingesting sources into a wiki, querying wiki knowledge, linting wiki quality, 'add to wiki', 'what do I know about', or any mention of 'LLM wiki' or 'Karpathy wiki'.

AI, LLMs & AgentsIn 1 list313

Seedance Prompt En

4.4k

dexhunter/seedance2-skill

Write effective prompts for Jimeng Seedance 2.0 multimodal AI video generation. Use when users want to create video prompts using text, images, videos, and audio inputs with the @ reference system. Covers camera movements, effects replication, video extension, editing, music beat-matching, e-commerce ads, short dramas, and educational content.

AI, LLMs & Agents356

Narrator AI CLI

3.1k

NarratorAI-Studio/narrator-ai-cli-skill

AI 电影/短剧解说视频自动生成（AI 解说大师 CLI Skill）。当用户需要创建电影解说视频、短剧解说、影视二创、AI 配音旁白视频、film commentary、video narration、drama dubbing、movie narration 时触发。内置电影素材库、BGM、多语种配音、解说模板。通过 narrator-ai-cli 命令行实现：搜片→选模板→选 BGM→选配音→生成文案→合成视频的全流程自动化。CLI client for Narrator AI video narration API.

AI, LLMs & Agents406

Asd Ste100

5.9k

danyuchn/asd-ste100-skill

Use when English text must be parsed without a human to resolve ambiguity — tool descriptions, error messages, inter-agent instructions, system prompts, status reports — and misreading has a real cost, or when text reads as dense, hedged, or easy to misparse. Triggers: disambiguate, STE100 rewrite, apply Simplified Technical English, plain-language rewrite, controlled-language rewrite, rewrite so an agent cannot misread this. Not for creative or marketing copy.

AI, LLMs & Agents340

Browse the most popular Claude skills, all ai, llms & agents skills, or write your own with the skills generator.