Being Human

io.github.Syedomershah99由 GitHub 账号 Syedomershah99 发布v0.1.2更新于 Oct 28, 2026

Learns your writing voice from your own prompt history, then makes the model write like that.

发布方自己的服务:TAP

概览

AI 生成的概览

从你自己的提示历史中学习个人写作风格,并输出风格规则,让助手写得更像你。

可以做什么
  • 从 Claude Code 历史、ChatGPT 导出文件或你自己的文件中采集并测量你的写作,可选地与模型一侧的对话做对比。
  • 返回带具体数字的风格规则,以及通过 log-odds 对比得出的个人化避用词表。
  • 检查草稿中的通用 AI 痕迹、节奏与结构。
  • 用基于你自己语料的作者距离,验证草稿在统计上是否出自你手。
  • 将风格档案导出给 ChatGPT、agents、Cursor、系统提示或 JSON,并生成可分享的声纹卡片。
  • 每个工具在返回自身结果的同时都会附带风格规则,不会只给分数而不给目标。
适用场景
适合希望助手草稿读起来像自己而非通用 AI 文风的人,也适合想要一个本地、纯算术的方式判断文本是否出自本人或属于套话。最适合已有大量既有写作可供学习的人。
运行要求
以本地 stdio 进程运行,从 PyPI 安装 being-human,用 uvx 或 being-human-mcp 命令启动。需要 Python 3(仅标准库,无依赖)以及本地语料:Claude Code 历史、ChatGPT 导出或你自己的文件。无网络请求、账号、API 密钥或环境变量。
安装前请注意
它会读取你已有的提示历史和写作文件,只应指向你愿意在本地处理的材料。采集时会脱敏密钥,档案目录也已加入 gitignore,但语料本身仍然敏感,尽管声纹可以安全分享。工具会向模型返回指导性文本,README 也指出没有任何机制强制模型调用它们。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Being Human,将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。

其他 MCP 客户端

参照 仓库 中的启动说明。

README

being-human

Your assistant writes like a press release because it doesn't know who you are. This tells it.

being-human reads the prompts you've already typed, measures how you actually write, and hands the model a description of your voice specific enough to follow. It also builds a list of words to avoid, computed by contrast against the model's own output, so it's personal to you rather than borrowed from a blog post about ChatGPT tells.

Local, stdlib Python 3, no dependencies, no network calls.

[being-human demo]

my own writing, measured by this tool. every bar is one real sentence. make your own with being-human card.


quickstart

bash
pip install being-human
being-human learn --name "Your Name"     # harvest your writing, measure itbeing-human check draft.md               # AI tellsbeing-human verify draft.md              # is it statistically you?being-human export --target chatgpt      # or agents, cursor, system, jsonbeing-human card                         # draw your voiceprint as a shareable svg

No Claude Code history? Point it at a ChatGPT export or your own files:

bash
being-human harvest --source chatgpt --path conversations.json --out .being-human/ --contrastbeing-human harvest --source files --path ~/Documents/writing --out .being-human/

what comes out

Instructions with numbers attached, because "be concise" is not actionable and "your median message is 12 words" is:

- Write "i" lowercase. They do, 83% of the time.- Vary sentence length hard. Theirs average 9 words but deviate 8, up to 81.- No exclamation marks. Rate is 0.28 per 1k.- Don't hedge. Hedge rate is 1.1 per 1k -- they assert.

the contrast trick

Your transcripts contain both sides of every conversation. So every word can be scored by log-odds: how much more likely you are to use it than the model is, on the same topics, in the same threads. The top of that list is your voice. The bottom is the model's.

Mine came back actually, clean, rather, genuinely, exactly, roughly. None of those appear on any published list of AI words. They're the specific tics this model has when talking to me. Yours will differ, because the contrast is against your writing.

two checks, not one

They catch different failures, and passing one says nothing about the other:

checkis this generic AI writing? tells, rhythm, structure
verifyis this you? authorship distance, Burrows's Delta against a length-matched null resampled from your own corpus

Measured on my own corpus:

authorshipslop
my own writing6th pct, me99/100
generic LinkedIn slop19th pct, me?0/100
a README by someone else94th pct, not me97/100

Row two is the catch: slop has unremarkable grammar, so authorship distance passes it. Row three is clean prose with no tells that is plainly a different hand. The second case is the one nobody else tests for, and it's the one that bites.

Held-out chunks of real writing sit at the 49th to 60th percentile, with a 1-6% false-positive rate above p97. Thresholds are personal. Em dashes get flagged against your measured rate rather than a universal rule, which is why this doesn't punish people who genuinely use them.

mcp server

bash
claude mcp add being-human -- being-human-mcp

Seven tools, two resources, two prompts. Configs for Claude Desktop, Cursor, Codex and Zed are in mcp/README.md.

MCP tools are model-invoked and nothing forces a model to call one, which is awkward when the payload is instructional. So every tool returns the voice rules alongside its own result. There's no path through the server that hands back a score without the target attached.

There's a Claude Code plugin too, via /plugin marketplace add Syedomershah99/being-human. It adds /being-human:learn, :check, :write, :verify, :export, and a hook that grows the corpus as you type.

privacy

Everything is local. Grep for urllib, requests or socket and you'll find nothing. Secrets are redacted at harvest time before anything touches disk, and .being-human/ is gitignored. The voiceprint is safe to share. The corpus it was built from is not.

prior art

This space got crowded in 2026. inside-lago framed the thesis, slop-guard is a mature slop linter, idiolect is architecturally closest, and writer-persona had the best idea of the lot: validate the profile with a backtest instead of trusting it.

Two things here I couldn't find elsewhere. The user-vs-model contrast, which derives the avoid-list from both sides of your own transcripts rather than from a curated list. And a deterministic impostor test. writer-persona uses an LLM judge across 8 axes; this is arithmetic, so it costs nothing and returns the same answer every time.

Burrows's Delta and log-odds with a Dirichlet prior are both established methods. The claim is the application and the calibration, not the statistics.

license

MIT. The tell list is data/slop-lexicon.json. If you keep seeing something that isn't in there, send a PR.

mcp-name: io.github.Syedomershah99/being-human

来源:README.md,提交 9826b30

工具

0
SourceWeft 尚未在沙箱中启动过该软件包,因此其工具未知。

版本历史

1
  1. v0.1.2最新Sep 16, 2026

评分

已满足适用检查项的 50%

该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。

  • 近 30 天内可连通25 分

    请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。

  • 已声明认证方式10 分

    请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。

  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分

其他 Human 集成