Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add taxueseek/say-it-human --skill humanize-aigit clone --depth 1 https://github.com/taxueseek/say-it-humanWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/taxueseek/say-it-human/humanize-ai)<a href="https://agentmods.dev/skills/taxueseek/say-it-human/humanize-ai"><img src="https://agentmods.dev/badge/skills/taxueseek/say-it-human/humanize-ai/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/taxueseek/say-it-human/humanize-ai"><img src="https://agentmods.dev/badge/skills/taxueseek/say-it-human/humanize-ai.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00229 | $0.05261 |
| Opus 5 | $0.00114 | $0.02631 |
| Sonnet 5 | $0.00046 | $0.01052 |
| Haiku 4.5 | $0.00023 | $0.00526 |
Grade A, and why
humanize-ai scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 11d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
This is a copy
89% identical to smart-search — 393 lines differ, which has more behind it and is treated as the original. This page carries a canonical link to it rather than competing with it.
How it starts
The opening of the file, as written. The whole thing — 344 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AI 味检测与消除
AI 味不是「写得不好」,是「读者感觉到背后没有真人」。但去 AI 味 ≠ 好内容。 花时间去 AI 味不如花时间把事情搞清楚。
文字洁癖(taxue):AI 味不是 AI 的问题,是用套话掩盖不确定的思考。人话三要素:带立场、带偏见、带情绪。每删一个套话,都要问:我到底想说什么?如果删完发现原文也没什么真东西,那问题不在 AI 味,在内容本身。
执行原则:
- 默认只诊断不改。识别AI特征是帮人看清自己的文字,不是帮人伪装成人类。改写必须用户确认。
- 去 AI 味 ≠ 好内容。如果检测结果很干净,直接告诉用户。不要为了输出报告而硬找问题。
- 改写必须基于用户的偏好。每个 AI 特征背后都有一个本来想达成的目的。追问那个目的,而不是直接替换。
- 见感思行是正面的改写目标(taxue)。去 AI 味不只是删坏东西,更是让文字回到「见(观察)→ 感(感受)→ 思(思考)→ 行(行动)」的自然结构。
三种改写方式(用户选):
- 自动替换:检测 → 直接改 → 质量验证 → 迭代
- 追问驱动:检测 → 逐条追问写作意图 → 用户回答后给修改方向
- 混合:自动替换 + 对关键特征追问意图
怎么运作:先判断AI味有多重,再选策略——轻微改几处、中等改一段、严重整篇重写、致命直接拦下。改完检查质量,不过关就回去重来,最多3轮。
怎么一步步处理
先判断内容类型(推文/公众号/学术),确定AI味的判断标准。然后快速扫一遍,数数AI味关键词命中几个。
- 命中少于5处 → 直接改标记的位置
- 命中5处以上 → 进入全量诊断,先找根因(3个根因能同时引发多种表现),再验证独立信号
改之前先识别作者的声音——口头禅、动情处、不完美痕迹,这些标为保护区一字不动。
遇到疑似AI味的句子,走决策树判断该不该删:在保护区?密度够?换话题还成立?是喘息句?删了会失温度?
改写时先清除明显废话,再逐段处理,最后检查人话三要素(偏见、情绪、立场)。
改完过检查:有没有编造?核心观点还在吗?有没有引入新的AI味?A/B级交付,C/D级回去重写。致命浓度直接拦住,不让改。
先判断内容类型
不同体裁对「规范表达」的容忍度不同。推文允许不完美,公众号保留结构感,学术文体别误伤。
| 体裁 | AI 味敏感度 | 处理原则 |
|---|---|---|
| 个人随笔/推文/短视频文案 | 高 | 允许不完美,允许跑题,允许情绪化 |
| 公众号/专栏文章 | 中 | 保留结构感,但打破模板化 |
| 商业报告/学术文体 | 低 | 规范表达不是 AI 味,别误伤 |
体裁判断只在有明显特征时做——大部分内容默认为公众号/专栏类型。详细判断标准 → references/extended-patterns.md。
快速扫描 + 全量诊断
快速扫描
扫一遍全文,数数这些关键词命中几个:
- 结构胶水:「首先…其次…最后…」「值得注意的是」「综上所述」
- 空洞词:「至关重要」「不可或缺」「深度赋能」「显著提升」
- 模板句:「在当今…背景下」「随着…的发展」
- 口水话:「如流星划过」「时不我待」
- 交流腔:「感谢您的」「期待与您」「让我们一起」
命中少于5处 → 直接改标记的位置。命中5处以上 → 进入全量诊断。
全量诊断
19种AI味模式,有3个根因能同时引发多种表现:
- 语气过于中性/权威 → 导致结构对称、句式整齐、RLHF模板化
- 套话与空洞修饰 → 导致概括语言多、情绪词空洞、口水话
- 翻译腔 → 导致RLHF模板化、虚假范围
先找根因,解决一个就能让多种表现同时消失。剩下的独立问题(缺乏细节、拔高、否定式煽情等)单独处理。不用每次过全部19种。
19 维速查表
每个维度的详细识别信号、修法、边界说明 →
references/extended-patterns.md
| # | 来源 | 识别信号 | 修法 |
|---|---|---|---|
| 1 | 结构对称 | 三段式排比、段落长度一致、每段总结句 | 打破对称,长短不一 |
| 2 | 概括语言多 | 「总的来说」「综上所述」、抽象词替代场景 | 用具体数字/人物替代 |
| 3 | 情绪词空洞 | 「令人深思」「至关重要」无细节 | 删情绪词,换具体细节 |
| 4 | 句式整齐 | 「不仅…而且…」大量重复、句子长度一致 | 长句后接极短句,换口语 |
| 5 | 缺乏细节 | 没时间地点人物、「研究表明」不说哪项 | 加时间、数字、具体描述 |
| 6 | 语气权威 | 整篇没有「我不确定」、像教科书 | 加入不确定、偏见、犹豫 |
| 7 | 过度拔高 | 「历史性时刻」「分水岭」「里程碑」 | 保留事物本来大小 |
| 8 | 否定式煽情 | 「不仅仅是 X,而是 Y」强行升华 | 直接说事实 |
| 9 | 虚假范围 | 「从 X 到 Y」硬凑、「涵盖多个领域」 | 删范围词,说做了什么 |
| 10 | 翻译腔 | 被动语态多、中英混杂、名词化堆叠 | 被动改主动,短句化 |
| 11 | 平台套路 | 小红书体/抖音体/知乎体全套照搬 | 像那个人,但不像模板 |
| 12 | RLHF 模板化 | 虚词密度异常、句子长度方差过低 | 故意打断流畅,加入停顿 |
| 13 | 格式型痕迹 | 破折号/粗体每段都有、表情符号、弯引号 | 克制格式,用直角引号 |
| 14 | 交流型痕迹 | 谄媚语气、免责声明、填充短语、过度限定 | 平等语气,直接说内容 |
| 15 | 口水话修辞 | 比喻连篇、空洞号召、套话、形容词堆砌 | 用数字替代比喻,删套话 |
| 16 | 套话修饰 | 「至关重要」密集、「在当今…」开头、万能句式 | 删掉或用数据替代 |
| 17 | 视角混乱 | 人称在「我」「我们」「你」之间漂移 | 确定稳定视角,全篇保持 |
| 18 | 伪逻辑连接 | 「因此」「所以」连接无因果关系的分句 | 检查因果,删假因果词 |
| 19 | 数字伪造 | 精确到小数但无来源、无上下文 | 有来源保留,无来源改模糊 |
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 11d ago First seen · 344 lines · 229 tokens per session scan A 0347a7d04734
humanize-ai is a skill published in the GitHub repository taxueseek/say-it-human (65 stars, last pushed 23d ago), licensed MIT. It adds 229 tokens to every session and 5,261 once invoked, about $0.0011 per session on Opus 5. A static security scan graded it A with 0 findings. It is 89% identical to smart-search, differing in 393 lines, and is treated as a copy.
Other skills, from other repositories
stop-slop-zh
A Blender animation guide for changing objects, cameras, lights, and other properties over time. It covers keyframes, motion curves, shape keys for morphs, drivers, and reusable animation actions.
AI去痕迹改写
A Chinese-language rewriting skill that removes 45 listed signs of AI-generated writing while preserving the original formatting.
AI去痕迹改写
A Chinese-language editor that removes writing patterns commonly associated with AI-generated text while keeping the original formatting unchanged.
OwnPen
A set of 45 Chinese writing rules for making text sound less machine-written while keeping its original formatting unchanged.
story-import
A tool for turning an existing novel into a structured writing project. It analyzes the book and organizes its characters, settings, plot plans, chapters, and tracking data for later writing.
story-long-analyze
A long-form fiction analysis workflow for breaking down a novel’s opening chapters, characters, pacing, turning points, relationships, and overall structure.