Writing Tone Clone — Personal Writing Style Replication
Clone someone's writing tone from sample text and produce a deployable voice skill.
What This Skill Does
Given one or more samples of someone's real writing, this skill:
- 1. Analyses the text across 8 linguistic dimensions
- Produces a documented voice profile
- Writes a deployable
[name]-voice SKILL.md the person can use immediately
The output is precise and specific — not generic labels like "conversational tone." Every finding is evidenced from the actual text.
Step 1 — Collect Samples
Ask for writing samples if not already provided. The more the better. Ideal sources:
- - Diaries or journals — most authentic; unguarded, natural voice
- Personal emails or messages — informal register
- Blog posts or social captions — may be more polished but still useful
- Voice note transcripts — captures spoken cadence
Minimum viable input: ~500 words of unedited personal writing.
If multiple sample types are available, use all of them and note which patterns appear across all types (most reliable) vs. only in one context.
Step 2 — Run the Analysis
Analyse the samples across these 8 dimensions. For each dimension, quote directly from the text as evidence.
Dimension 1 — Sentence Structure & Length
- - What is the typical sentence length? (Short/punchy vs. long/flowing)
- Do sentences vary in length or stay consistent?
- Are sentences mostly simple, compound, complex, or mixed?
- Does the writer use fragments or run-ons intentionally?
Dimension 2 — Voice & Register
- - Formal or casual? Where on the spectrum?
- Does the writer address the reader directly ("you")? How often?
- First person throughout, or does it shift?
- Is there self-awareness / self-commentary in the writing?
Dimension 3 — Humour & Tone Markers
- - Is there humour? What kind — dry, slapstick, self-deprecating, absurdist?
- Are there ironic asides or understatements?
- Where does the writer make jokes and how are they constructed?
- Is the humour visible in punctuation choices (brackets, dashes, ellipsis)?
Dimension 4 — Rhythm & Pacing
- - Does the writing accelerate into action scenes and slow for reflection?
- Are there deliberate one-line beats for emphasis?
- How are lists used — and do they feel clinical or casual?
- Does punctuation drive rhythm (em dashes, semicolons, commas)?
Dimension 5 — Vocabulary & Word Choice
- - What is the reading level? (Grade school / conversational / formal?)
- Does the writer avoid jargon, or use it naturally?
- Are there recurring words or phrases?
- Are there conspicuous word choices that feel very "them"?
- What words would they never use?
Dimension 6 — Emotional Register
- - How openly does the writer express emotion?
- Are feelings stated directly or implied through action/observation?
- How does the writer handle vulnerability — lean into it or downplay it?
- Is positivity, negativity, or neutrality dominant?
Dimension 7 — Structural Habits
- - How does the writer open paragraphs and entries?
- How do they close or trail off?
- Are transitions used? What kind?
- Do they use bullet points, or resist structure?
- How much scene-setting vs. getting-to-the-point?
Dimension 8 — What They Don't Do
This is often the most useful dimension. Note:
- - Topics, words, or phrases they avoid entirely
- Rhetorical devices they never use (metaphors? hyperbole? statistics?)
- Sentence types they don't write
- Emotional registers they don't express
- Things a generic AI would write that this person never would
Step 3 — Produce the Voice Profile
Compile the 8 dimensions into a single Voice Profile document. Format:
CODEBLOCK0
Step 4 — Write the Voice Skill
Using the Voice Profile, write a deployable SKILL.md for this person's voice. The skill should:
- - Have a clear description triggering on "write in [name]'s voice", "sound like [name]", etc.
- Load the voice profile as a reference document
- Instruct the agent to follow the Quick Reference Rules strictly
- Include 2-3 worked examples showing the voice applied to different content types
- Include the "What To Avoid" section prominently
Save to: skills/[name]-voice/SKILL.md
Save the voice profile to: skills/[name]-voice/references/voice-profile.md
Notes on Quality
- - Evidence over assertion. Never say "conversational tone" without quoting the text that shows it.
- Specific over general. "Uses em dashes to insert dry asides" beats "informal punctuation."
- Negative space matters. What someone doesn't write is as defining as what they do.
- Context-dependent. Note if a pattern only appears in certain contexts (e.g. humour only in personal writing, not professional).
- Iterate if possible. If the person can provide feedback on a first attempt, use it to refine the profile.
写作语调克隆 — 个人写作风格复制
从样本文本中克隆某人的写作语调,并生成一个可部署的语音技能。
该技能的功能
给定某人真实写作的一个或多个样本,该技能将:
- 1. 从8个语言维度分析文本
- 生成一份有据可查的语音档案
- 编写一份可部署的[姓名]-voice SKILL.md文件,供该人立即使用
输出结果精准且具体——而非对话式语调这类泛泛标签。每项发现均源自实际文本的佐证。
第一步 — 收集样本
如尚未提供写作样本,请主动索取。样本越多越好。理想来源包括:
- - 日记或日志 — 最真实;未经修饰、自然的表达
- 个人邮件或消息 — 非正式语体
- 博客文章或社交媒体文案 — 可能更精炼,但仍具参考价值
- 语音笔记转录稿 — 捕捉口语节奏
最低有效输入:约500字未经编辑的个人写作内容。
如有多类样本可用,请全部使用,并标注哪些模式在所有类型中均出现(最可靠),哪些仅出现在特定语境中。
第二步 — 执行分析
从以下8个维度分析样本。针对每个维度,直接引用原文作为佐证。
维度一 — 句子结构与长度
- - 典型句子长度如何?(短促有力 vs. 长句流畅)
- 句子长度是否多变,还是保持稳定?
- 句子主要为简单句、并列句、复合句,还是混合使用?
- 作者是否有意使用句子片段或流水句?
维度二 — 语气与语体
- - 正式还是随意?处于何种程度?
- 作者是否直接称呼读者(你)?频率如何?
- 通篇使用第一人称,还是有所变化?
- 文中是否存在自我意识或自我评论?
维度三 — 幽默与语气标记
- - 是否存在幽默?属于何种类型——冷幽默、滑稽、自嘲、荒诞?
- 是否有讽刺性的旁白或轻描淡写?
- 作者在何处制造笑点,又是如何构建的?
- 幽默是否体现在标点符号的选择上(括号、破折号、省略号)?
维度四 — 节奏与韵律
- - 写作是否在动作场景中加速,在反思时放缓?
- 是否有刻意使用单行节奏来强调?
- 列表如何使用——感觉像临床记录还是随意列举?
- 标点符号是否驱动节奏(长破折号、分号、逗号)?
维度五 — 词汇与用词选择
- - 阅读难度如何?(小学水平/对话式/正式?)
- 作者是否避免使用术语,还是自然运用?
- 是否存在反复出现的词汇或短语?
- 是否有明显带有个人特色的用词选择?
- 哪些词汇是作者绝不会使用的?
维度六 — 情感基调
- - 作者表达情感的开放程度如何?
- 情感是直接陈述,还是通过行动/观察间接暗示?
- 作者如何处理脆弱感——直面还是淡化?
- 积极、消极还是中立情绪占主导?
维度七 — 结构习惯
- - 作者如何开启段落和章节?
- 如何收尾或逐渐淡出?
- 是否使用过渡词?使用何种类型?
- 是否使用项目符号,还是抗拒结构化?
- 场景铺垫与直入主题的比例如何?
维度八 — 作者不会做的事
这通常是最有用的维度。请记录:
- - 完全回避的话题、词汇或短语
- 从不使用的修辞手法(隐喻?夸张?数据统计?)
- 不写的句子类型
- 不表达的情感基调
- 通用AI会写但此人绝不会写的内容
第三步 — 生成语音档案
将8个维度的分析结果整合成一份语音档案文档。格式如下:
[姓名] 语音档案
概述
[2-3句话总结该语音特点——让陌生人能立即识别]
8个维度
[每个维度作为带标题的章节,包含发现结果+引用佐证]
快速参考规则
[10-15条要点:用该语音写作时最重要的注意事项]
应避免的内容
[明确列出会破坏该语音的事项]
示例短语
[5-10个明显属于此人的短语或句子类型]
第四步 — 编写语音技能
基于语音档案,为此人的语音编写一份可部署的SKILL.md文件。该技能应:
- - 包含清晰的描述,触发词如用[姓名]的语音写作、听起来像[姓名]等
- 将语音档案作为参考文档加载
- 指示智能体严格遵守快速参考规则
- 包含2-3个示例,展示该语音在不同内容类型中的应用
- 突出显示应避免的内容部分
保存至:skills/[姓名]-voice/SKILL.md
将语音档案保存至:skills/[姓名]-voice/references/voice-profile.md
关于质量的说明
- - 证据优先于断言。 切勿在未引用文本佐证的情况下说对话式语调。
- 具体优先于笼统。 使用长破折号插入冷幽默旁白优于非正式标点。
- 留白同样重要。 作者不写的内容与所写内容同样具有定义性。
- 语境相关。 注意某模式是否仅出现在特定语境中(例如,幽默仅出现在个人写作中,而非专业写作)。
- 如有可能,迭代优化。 如果作者能对初稿提供反馈,请据此完善档案。