Back to Details

ig-humanizer

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Instagram Humanizer V3

Instagram Humanizer V3

Rewrites any caption or carousel slide text to remove the AI tells that human readers notice, and audits a finished caption against the 2026 Instagram checklist. Based on Wikipedia's "Signs of AI writing" taxonomy, the 2025-2026 stylometry literature, our own Instagram caption corpus (n=284, prevalence only), and Instagram-specific patterns (the first-125 fold, the lowercase-casual caption register, sized hashtags, sends-and-saves structure). V3 (2026-09): recalibrated on 2026 evidence. Vocabulary is scored by density, em dashes are capped instead of banned (and the cap is generous here: 29% of human captions use one), forced rhythm is now a tell instead of a fix, and there is an over-correction guard.
What this skill does not do: it does not make text "pass" GPTZero, Pangram, Turnitin or Originality. Those are trained classifiers keyed on the instruction-tuning style signature; prompt-style "sound like a real person" rewrites are caught 92-95% of the time, and light mechanical rewriting raises detectability. On caption-length text (under 300 words) detector scores are noise. The real value is elsewhere: expert human readers cite vocabulary (53%) and sentence structure (36%) as what gives AI text away, and on Instagram a caption that reads as a brand account earns neither the save nor the send. This skill removes what those readers react to.
重写任何 caption 或 carousel slide 文案，以去除人类读者会注意到的 AI tell（破绽），并依据 2026 年 Instagram checklist 审核已完成文案。依据 Wikipedia 的 “Signs of AI writing” 分类体系、2025-2026 年文体测量学（stylometry）文献、我们自己的 Instagram caption 语料库（n=284，仅统计出现率），以及 Instagram 特有模式（first-125 fold、lowercase-casual caption register、数量受控的 hashtags、发送与收藏结构）。
V3（2026-09）： 根据 2026 年证据重新校准。词汇按密度评分；em dash 从禁用改为限量（这里的上限较宽松：29% 的人类 caption 会使用一个）；刻意制造的节奏变化现在被视为 tell，而不再作为修正手段；并加入了过度修正保护。
这个 skill 不做什么： 它不会让文本“通过” GPTZero、Pangram、Turnitin 或 Originality。这些工具训练的分类器以 instruction-tuning 风格签名为依据；提示词式的“听起来像真人”改写有 92-95% 的概率会被识别，而轻度机械改写反而会提高可检测性。对于 caption 长度的文本（少于 300 个词），检测器分数基本是噪声。真正的价值在别处：专业的人类读者认为，词汇（53%）和句子结构（36%）最容易暴露 AI 文本；在 Instagram 上，一段读起来像品牌账号的 caption 既不会被收藏，也不会被转发。这个 skill 会去除这些读者会反感的特征。

What changed in V3

V3 的变化

Evidence tier in brackets: [strong] = replicated across 2+ independent 2025-2026 studies or our own corpus; [vendor] = single platform or vendor dataset; [weak] = one study or expert-panel report.
  • Vocabulary moved from a delete-list to density scoring. The 2023-24 words (delve, tapestry, realm, journey) are decaying as humans avoid them [strong]. The durable 2026 markers are common words (significant, crucial, notably, comprehensive, insights, robust, leverage, foster, landscape, nuanced, streamline, elevate) plus grammar: nominalisations and "-ing" clause openers at 5.3x the human rate [strong]. AI vocabulary appears in 10% of human captions in our corpus [strong], so one marker in a paragraph is not a verdict. Three is.
  • Em dash is no longer a tell. GPT-5.4 emits 1.43 per 1,000 words, below the 3.23 human baseline [strong], and 29% of human Instagram captions contain one [strong: corpus]. A caption's em dash is never a tell on its own, and a blanket ban over-sterilises captions. New rule: cap at about 1 per 100 words (1-2 per caption), replace only the excess with a comma, colon, parentheses or a rewrite. Never a period (a split dash stacks fragments). Zero em dashes across a long caption is its own tell now.
  • Forced burstiness is the #1 2026 tell, not the fix. Mechanical long/short alternation is a learnable humanizer fingerprint [weak], and "Short. Punchy. Done.", "No X. No Y. Just Z.", one-word lines for drama and "The result?" reveals are the current top reader-cited tells [strong]. Captions are mid-length, so Pass 2 is an anti-uniformity guard only: fix a paragraph that reads machine-flat, never manufacture variance.
  • Rule of three is still a tell, at density. Tricolon runs at 2x expert-human rate across 2026 frontier models [strong], and 23% of human captions contain one [strong: corpus]. So one natural triple with concrete items stays. Stacked, perfectly parallel or hollow triads and a third triad in a caption get scrubbed.
  • Fingerprint injection was half wrong. Named entities and concreteness are supported [strong]; an odd-precision number with a referent in the first 125 chars is the strongest hook. Bare numbers are not a discriminator, and inserted hedges and confessions backfire: performed hesitancy is 2x more common in LLM text, and sincerity announcements ("let me be honest", "real talk", "POV:" on something that is not a POV) are a named 2026 tell [strong]. Pass 3 asks for a flat, dated, uncomfortable fact instead.
  • Over-correction guard. Humanizer output has its own fingerprint [weak]. Pass 4 checks whether Passes 1-3 introduced the very patterns they were meant to remove. Edits are proportional to real problems. When in doubt, leave it.
证据等级用方括号标注：[strong] = 在 2 项以上独立研究中得到复现，或得到我们自己的语料库支持；[vendor] = 仅来自单一平台或厂商的数据集；[weak] = 来自一项研究或专家小组报告。
  • 词汇从删除清单改为密度评分。 2023-24 年的词（delve、tapestry、realm、journey）正在失宠，因为人类正在避开它们 [strong]。2026 年持久存在的标记是一些常见词（significant、crucial、notably、comprehensive、insights、robust、leverage、foster、landscape、nuanced、streamline、elevate），以及语法特征：名词化和以 “-ing” 从句开头的使用率是人类写作的 5.3 倍 [strong]。在我们的语料库中，AI 词汇也出现在 10% 的人类 caption 中 [strong]，因此一段中出现一个 marker 还不能下定论，出现三个才是。
  • Em dash 不再是 tell。 GPT-5.4 每 1,000 个词输出 1.43 个 em dash，低于人类 3.23 个的基线 [strong]，而且 29% 的人类 Instagram caption 含有一个 em dash [strong：corpus]。单独出现的 em dash 从来不是 tell，全面禁用只会让 caption 显得过度净化。新规则：每 100 个词最多约 1 个（每条 caption 1-2 个），只将超出的部分替换为逗号、冒号、括号或重新改写。绝不能换成句号（拆分会形成堆叠的残句）。如今，长 caption 中完全没有 em dash 本身也是一种 tell。
  • 刻意制造的句长起伏是 2026 年最大的 tell，而不是修正手段。 机械式长短句交替已成为可学习的 humanizer 指纹 [weak]；“Short. Punchy. Done.”、“No X. No Y. Just Z.”、为制造戏剧感而使用单词成行的句子，以及 “The result?” 式揭示，是当前读者提及最多的 tell [strong]。Caption 通常为中等长度，因此 Pass 2 仅作为反均匀性保护：修正读起来机械、僵硬的段落，绝不人为制造句长差异。
  • 三项法则仍然是一种 tell，但要看密度。 2026 年前沿模型使用三项并列句式的频率是专家人类写作者的两倍 [strong]，且 23% 的人类 caption 中含有一处 [strong：corpus]。因此，一组包含具体项目、表达自然的三项并列可以保留。堆叠、完全平行或空洞的三段式，以及 caption 中的第三组三段式都应被清理。
  • 指纹注入的思路此前有一半是错误的。 具名实体和具体性有证据支持 [strong]；位于 first 125 个字符内、带有明确指代对象的异常精确数字，是最强的 hook。裸露的数字无法区分文本来源，插入 hedge 和自白反而会适得其反：表演式犹豫在 LLM 文本中的出现率是专家人类文本的两倍，而真诚宣告（“let me be honest”、“real talk”，或在与 POV 无关的内容上使用 “POV:”）是公认的 2026 年 tell [strong]。Pass 3 会改为要求一个平铺直叙、有日期且令人不适的事实。
  • 过度修正保护。 Humanizer 输出本身也有指纹 [weak]。Pass 4 会检查 Pass 1-3 是否恰好引入了原本要消除的模式。所有编辑都应与实际问题相称。不确定时，保留原样。

When to use

何时使用

  • Before publishing any AI-drafted caption or carousel (rewrite mode)
  • Pre-publish review of a finished caption (audit mode, see 
    sub-skills/post-audit.md
    )
  • When a caption feels off and you cannot pinpoint why
  • 发布任何由 AI 起草的 caption 或 carousel 之前（改写模式）
  • 对已完成 caption 进行发布前审核（audit 模式，参见 
    sub-skills/post-audit.md
    ）
  • 当一段 caption 感觉不对，却无法确定具体原因时

Input

输入

Any text: a caption, the on-image text of a carousel, or a Reel caption. Optional: target voice samples (the user's past captions).
任意文本：一条 caption、carousel 的图上文字或 Reel caption。 可选：目标 voice 样本（用户过去的 caption）。

Output

输出

  • Rewritten text with AI tells removed
  • A diff showing what changed and why
  • Caption char count (flagging if the hook spills past the first 125 chars)
  • Per-paragraph tell density (markers per paragraph; 3+ triggered a rewrite)
  • Reader-read confidence: "reads human", "mixed", "reads AI" (a reader-tell estimate, not a detector score)
  • 移除 AI tell 后的改写文本
  • 展示改动内容及改动原因的 diff
  • Caption 字符数（如果 hook 超过前 125 个字符则进行标记）
  • 每段 tell 密度（每段的 marker 数量；达到 3 个即触发改写）
  • 读者感置信度：“reads human”、“mixed”或“reads AI”（这是对读者 tell 的估计，不是 detector 分数）

Modes

模式

bash
undefined
bash
undefined

Default: scrub AI tells (forensic + strict) and fix Instagram-format issues

默认：清理 AI tell（forensic + strict），并修正 Instagram 格式问题

ig-humanizer <text>
ig-humanizer <text>

Forensic only - minimum touch, just kill model leakage

仅 forensic——最少改动，只清除模型泄漏

ig-humanizer --mode forensic <text>
ig-humanizer --mode forensic <text>

Audit - detection-only pass-fail review, no rewrite

Audit——仅检测的通过/失败审核，不改写

Runs the 2026 Instagram checklist: first-125 hook, caption length, sized

运行 2026 年 Instagram checklist：first-125 hook、caption 长度、

hashtags, emoji limits, CTA quality, media reminder.

数量受控的 hashtags、emoji 限制、CTA 质量和 media 提醒。

Returns Blockers + Warnings + suggested fixes. See sub-skills/post-audit.md.

返回 Blockers + Warnings 及建议修复。参见 sub-skills/post-audit.md。

ig-humanizer --mode audit <text>
ig-humanizer --mode audit <text>

Profile - build/update the user's Voice & Brand Profile. See the section below.

Profile——建立或更新用户的 Voice & Brand Profile。参见下方章节。

ig-humanizer --mode profile
undefined
ig-humanizer --mode profile
undefined

The four passes

四个 Pass

Pass 1 - SCRUB (score, then delete or replace)

Pass 1 - SCRUB（评分，然后删除或替换）

Apply the tiered catalogs in 
references/scrub-rules.md
. The unit of judgement is the paragraph (or slide), not the word: count markers per paragraph, rewrite the paragraph at 3+, leave a single marker alone unless it is a reveal bridge, negative parallelism, a sincerity marker, or forensic leakage.
  • Forensic (always on): real model leakage no human types. AI tool markers (oaicite, contentReference, turn0search0), knowledge-cutoff disclaimers ("As of my last update"), template blanks ([Your Name]), and em dashes above the cap (more than about 1 per 100 words).
  • Strict (default on): what readers react to. The durable 2026 vocabulary set scored by density (significant, crucial, notably, particularly, comprehensive, insights, robust, leverage, foster, landscape, nuanced, streamline, elevate, empower), grammar markers (nominalisations, sentence-opening "-ing" clauses), the 2026 model-idiom layer (quietly, "X matters.", compound, "a signal", "the work", "built different", "let that sink in"), reveal bridges on a single hit ("The result?", "Here's what", "Stop X, start Y", "plot twist:"), all forms of negative parallelism, stacked or perfectly parallel triads and any third triad in a caption, phrase cleanups ("in today's fast-paced world", "game-changer", "level up", "dive in"), emoji storms, and dead closers ("what do you think?", "double tap if you agree").
  • Instagram-format scrubs (always apply): first-125 hook that stands alone, sized hashtags, emoji limits, carousel slide-1 promise, caption length.
应用 
references/scrub-rules.md
中按等级划分的目录。判断单位是 段落（或幻灯片），而不是单词：统计每段的 marker 数量，达到 3 个时改写整段；只有一个 marker 时保持不动，除非它是 reveal bridge、否定式平行结构、真诚标记或 forensic 泄漏。
  • Forensic（始终开启）：清除真人不会输入的模型泄漏。包括 AI 工具标记（oaicite、contentReference、turn0search0）、知识截止日期免责声明（“As of my last update”）、模板占位符（[Your Name]），以及超过上限的 em dash（每 100 个词多于约 1 个）。
  • Strict（默认开启）：清除读者会反感的特征。处理 2026 年持久存在的词汇集合并按密度评分（significant、crucial、notably、particularly、comprehensive、insights、robust、leverage、foster、landscape、nuanced、streamline、elevate、empower）、语法标记（名词化、以 “-ing” 从句开头）、2026 年模型惯用语层（quietly、“X matters.”、compound、“a signal”、“the work”、“built different”、“let that sink in”）、单次命中即需处理的 reveal bridges（“The result?”、“Here's what”、“Stop X, start Y”、“plot twist:”）、所有形式的否定式平行结构、堆叠或完全平行的三段式及 caption 中出现的第三组三段式、短语清理（“in today's fast-paced world”、“game-changer”、“level up”、“dive in”）、emoji 风暴和无效收尾（“what do you think?”、“double tap if you agree”）。
  • Instagram 格式清理（始终应用）：确保 first-125 hook 能够独立成立，控制 hashtags 数量和 emoji 数量，检查 carousel 第 1 页的承诺以及 caption 长度。

Pass 2 - RHYTHM (anti-uniformity guard only)

Pass 2 - RHYTHM（仅作为反均匀性保护）

Detectors do not score burstiness, and captions are mid-length (our corpus median is well under 200 words), so rhythm is neither a reach lever nor a fix. What readers notice is the mechanical-uniformity tell (every line the same length, machine-flat) and, worse, the staged variance that second-generation humanizers add. So Pass 2 has two jobs: fix rhythm only where it reads machine-flat, and remove manufactured variance everywhere. It never adds variance as a tactic.
  • Per paragraph: one genuinely long sentence next to a short one is fine and is what human variance looks like. Two or three mid-length sentences in a row are also fine. Edit only when every sentence in the paragraph runs the same length and reads flat, and then edit one sentence, not the paragraph.
  • Standalone fragments: at most 2 per caption, total. "every time." once is a voice quirk. Three in a caption is a pattern.
  • Banned outright (rewrite as full sentences): "The X? Y." reveals; "No X. No Y. Just Z."; "All the X. None of the Y."; "Simple. Effective. Easy." adjective stacks; one-word lines for drama ("Still." "Exactly."); pseudo- Socratic Q&A ("Why? Because..."); "Short. Punchy. Done." staccato runs. Fragment runs are the tell.
  • Layout is not rhythm. One idea per line with blank lines between them is native caption formatting and stays. Fragment-for-drama inside those lines is the tell. Keep the layout, fix the sentences.
  • Carousel slide text is short by design; never chop or pad a slide for rhythm. Never alternate long/short/long/short across slides or lines. That seesaw is the humanizer fingerprint.
The check is "does any paragraph read machine-flat, and did I add a staccato pattern", not a variance number.
Detector 不会对 burstiness 评分，而 caption 通常为中等长度（我们的语料库中位数远低于 200 个词），因此节奏既不是触达杠杆，也不是修正手段。读者真正注意到的是机械式均匀这一 tell（每行长度相同、读起来僵硬的机械感），更糟的是第二代 humanizer 添加的刻意句长差异。因此 Pass 2 有两项任务：只在内容读起来机械、僵顿时修正节奏，并在所有位置移除人为制造的差异。它绝不把句长变化当作一种策略来添加。
  • 按段落检查：一个真正较长的句子旁边有一个短句完全没问题，这正是人类写作的自然差异。连续两三个中等长度的句子也可以接受。只有当段落中每个句子的长度都相同且读起来僵平时才进行编辑，而且只修改一个句子，而不是重写整个段落。
  • 独立残句：每条 caption 总共最多 2 个。“every time.” 出现一次可以是个人 voice 的特征；一条 caption 中出现三个就成了一种模式。
  • 全面禁止（改写为完整句子）：“The X? Y.” 式揭示；“No X. No Y. Just Z.”；“All the X. None of the Y.”；“Simple. Effective. Easy.” 式形容词堆叠；为制造戏剧感而使用单词成行的句子（“Still.”、“Exactly.”）；伪苏格拉底式问答（“Why? Because...”）；“Short. Punchy. Done.” 式断奏堆叠。残句连续成串就是 tell。
  • 排版不等于节奏。每个想法单独一行，并在行间留空，是 Instagram caption 的原生格式，应予保留。在这些行内为制造戏剧感而使用残句才是 tell。保留排版，只修正句子。
  • Carousel slide 文本的设计本来就很短；绝不能为了制造节奏而切碎或注水。也绝不能跨 slide 或跨行制造长/短/长/短的跷跷板式变化。这种模式正是 humanizer 指纹。
检查标准是“有没有任何段落读起来机械、僵硬，以及是否添加了断奏模式”，而不是计算句长差异。

Pass 3 - ADD (human fingerprints)

Pass 3 - ADD（添加人类指纹）

Require where the content allows:
  • One odd-precision number WITH a named referent: who, what, when, or what it cost ("0 to 10k in 4 months, posting 3 times a week", not "grew fast" and not "10k"). A bare number is not a fingerprint; the referent carries the signal.
  • One named entity (real person, brand, date, tool)
  • One first-person concrete detail (what you saw, what it cost, what broke)
  • One specific, dated, uncomfortable fact stated flat, with no framing sentence before or after it. Not "real talk, this one hurt: I lost the client." Just "I lost my biggest client on 14 Feb." The fact carries the vulnerability. The frame turns it into performed sincerity, which readers now read as the tell.
  • The lowercase-casual register if the voice calls for it
Forbidden as openers or pivots (sincerity announcements, a named 2026 tell): "let me be honest", "I'll be real", "honestly?", "to be direct", "the honest version is", "real talk", "not gonna lie", "ngl", "can I be vulnerable for a second", "unpopular opinion:" as a preface to a popular one, "POV:" on something that is not a POV. Also forbidden as insertions: hedges the author did not write ("perhaps", "I might be wrong but", "it seems"). Performed hesitancy is 2x more common in LLM text than in expert human text; adding it makes the draft read more AI, not less.
If the input lacks these, ask the user for a number, name, or moment. Do not fabricate.
在内容允许时加入：
  • 一个带明确指代对象的异常精确数字：说明是谁、什么、何时或花费了多少（“0 to 10k in 4 months, posting 3 times a week”，而不是“grew fast”或“10k”）。裸露的数字不构成指纹，指代对象才携带信号。
  • 一个具名实体（真实人物、品牌、日期或工具）
  • 一个第一人称的具体细节（看到了什么、花费了多少、什么出了问题）
  • 一个具体、有日期且令人不适的事实，平铺直叙地陈述，前后不加任何框架句。不要写“real talk, this one hurt: I lost the client.”，只写“I lost my biggest client on 14 Feb.”。事实本身承载脆弱感；框架会把内容变成表演式真诚，而读者现在会将其视为 tell。
  • 如果 voice 本身需要，则采用 lowercase-casual register
禁止将以下内容用作开头或转折（这些是真诚宣告，是公认的 2026 年 tell）：“let me be honest”、“I'll be real”、“honestly?”、“to be direct”、“the honest version is”、“real talk”、“not gonna lie”、“ngl”、“can I be vulnerable for a second”，以及在与 POV 无关的内容上使用“POV:”。也禁止插入作者没有写过的 hedge（“perhaps”、“I might be wrong but”、“it seems”）。表演式犹豫在 LLM 文本中的出现率是专家人类文本的两倍；加入它只会让草稿显得更像 AI，而不是更像人。
如果输入内容缺少这些信息，应向用户询问数字、名称或具体时刻。不得编造。

Pass 4 - SELF-CHECK (over-correction guard)

Pass 4 - SELF-CHECK（过度修正保护）

Humanizer output has its own fingerprint. Before returning, re-read the result once and answer three questions:
(a) Did Pass 2 create staccato stacks, "The result?" reveal bridges, one-word lines, or a long/short/long/short seesaw? If yes, merge the fragments back into full sentences. (b) Did Pass 3 add a framed confession, a sincerity announcement, or a hedge the author never wrote? If yes, strip the frame and keep only the flat fact, or remove the insertion. (c) Did scrubbing flatten the author's voice: uniform tone, no reaction, no concrete detail left, every em dash gone, every triad gone, every emoji gone from a voice that uses them, the lowercase register capitalised? If yes, restore what the author had. Zero em dashes and zero triads in a long caption is a tell in its own right.
If any answer is yes, dial back rather than scrub harder. Edits must be proportional to real problems: a clean caption gets two or three touches, not a quota. When in doubt whether a pattern is the author or the model, leave it.
Humanizer 输出本身也有指纹。返回结果前，重新阅读一次并回答三个问题：
(a) Pass 2 是否创建了断奏堆叠、“The result?” 式 reveal bridge、单词成行的句子，或长/短/长/短的跷跷板式变化？如果是，将残句合并回完整句子。
(b) Pass 3 是否添加了带框架的自白、真诚宣告，或作者从未写过的 hedge？如果是，移除框架并只保留平铺直叙的事实，或者删除插入内容。
(c) 清理是否抹平了作者的声音：语气完全统一、没有任何反应、没有留下具体细节、所有 em dash 都被删除、所有三段式都被删除、本来使用 emoji 的声音被清空所有 emoji，或 lowercase register 被大写？如果是，恢复作者原有的特征。长 caption 中零 em dash、零三段式，本身也是一种 tell。
如果任何问题的答案为“是”，应减少修改，而不是加大清理力度。所有编辑必须与实际问题相称：一条干净的 caption 只需修改两三处，而不是为了凑数而编辑。当无法确定某个模式来自作者还是模型时，保留原样。

Non-negotiable rules

不可协商的规则

Global voice rules: see root 
SKILL.md
Voice rules. Additional skill-specific rules (V3):
  • Scrubbing is always in scope. When asked to humanize, de-AI, finalize, or publish a caption, run at least the forensic + strict passes before it ships. This holds when the user wrote the draft themselves, says they love it as-is, or is in a hurry. Author identity, "it's already good," and time pressure are never reasons to skip the scrub. The forensic + strict pass changes no meaning and takes seconds: run it, then ship. If a constraint truly forbids touching the text, say so explicitly and name every tell left in; the default is to scrub, not to wave it through.
  • Scrub proportionally. A pass that finds nothing changes nothing. Do not invent edits to justify the run, and do not report a detector score as the result; report the tells found and fixed.
  • Preserve the user's actual claim and meaning. "Preserve their voice" covers voice quirks and what they are claiming, NOT reveal bridges, staccato stacks, or a paragraph with 3+ vocabulary markers. Stripping those is not changing their voice; it is the job.
  • Never introduce facts that were not in the input. If a number is missing, ask.
  • Never introduce sincerity markers, hedges, or confessional frames. If the draft needs a vulnerable beat, ask for a dated fact and state it flat.
  • Keep the user's voice quirks (lowercase starts, 
    ..
    soft pauses, one em dash per ~100 words, one natural triad, their 1-3 intentional emoji).
  • Never promise detector results. If the user asks "will this pass GPTZero," answer honestly: nobody can promise that, and the score on a 150-word caption is noise.
  • Respect the surface: do not turn a carousel's slide text into a single caption or vice versa without flagging it.
  • Never present a caption as publishable without the media reminder (Instagram needs an image or video).
全局 voice 规则：参见根目录 
SKILL.md
中的 Voice rules。其他 skill 专属规则（V3）：
  • 清理始终在处理范围内。 当用户要求 humanize、de-AI、finalize 或发布 caption 时，发布前至少执行 forensic 和 strict 两个 pass。即使用户自己写了草稿、表示非常喜欢当前版本，或急于发布，也必须执行。作者身份、“已经写得很好”和时间压力都不是跳过清理的理由。Forensic + strict 不会改变含义，只需几秒钟：先执行，然后发布。如果某项约束确实禁止改动文本，应明确说明，并列出所有遗留的 tell；默认行为是清理，而不是直接放行。
  • 按比例清理。 没有发现问题的 pass 不做任何修改。不要为了证明执行过而编造编辑，也不要把 detector 分数作为结果；应报告发现并修复的 tell。
  • 保留用户实际表达的主张和含义。“保留他们的 voice”涵盖个人语言特征及其主张，但不涵盖 reveal bridge、断奏堆叠，或包含 3 个以上词汇 marker 的段落。去除这些内容不是改变作者的声音，而是这项工作的目的。
  • 绝不引入输入中不存在的事实。如果缺少数字，应进行询问。
  • 绝不引入真诚标记、hedge 或自白式框架。如果草稿需要一段脆弱表达，应询问一个有日期的事实，并平铺直叙地陈述。
  • 保留用户的 voice 特征（lowercase 开头、
    ..
    式停顿、每约 100 个词一个 em dash、一组自然三段式，以及用户有意使用的 1-3 个 emoji）。
  • 绝不承诺 detector 结果。如果用户问“会通过 GPTZero 吗”，应诚实回答：没人能够保证，而且 150 词 caption 的分数基本是噪声。
  • 尊重内容载体：不要在未标记的情况下，把 carousel 的 slide 文本合并成一条 caption，或反过来转换。
  • 如果没有 media 提醒，绝不能将 caption 呈现为可以直接发布；Instagram 需要图片或视频。

Instagram-specific tells this skill catches

这个 skill 能识别的 Instagram 特有 tell

  • A hook that needs the second line to make sense (the "more" fold eats it).
  • A caption opening with "Let's talk about..", "Here's the thing..", or "POV:" on something that is not a POV.
  • 20-30 hashtags crammed at the top instead of a 3-5 sized set.
  • Emoji sprinkled through every line (4+ per caption).
  • A bare title on carousel slide 1 instead of a promise + loop.
  • Stacked or hollow rule-of-three lists ("faster, cheaper, better"); one natural triple with concrete items is fine.
  • Engagement bait ("comment YES", "tag 3 friends", "double tap if").
  • Staccato stacks and one-word lines for drama (the humanizer fingerprint), or a paragraph where every line reads machine-flat.
  • 需要第二行才能理解的 hook（“more” fold 会吞掉 hook）
  • Caption 以“Let's talk about..”、“Here's the thing..”开头，或在与 POV 无关的内容上使用“POV:”
  • 在顶部塞入 20-30 个 hashtags，而不是使用 3-5 个数量受控的 hashtags
  • 每一行都散布 emoji（每条 caption 4 个以上）
  • Carousel 第 1 页只有裸露的标题，而不是“承诺 + 闭环”
  • 堆叠或空洞的三项法则列表（“faster, cheaper, better”）；一组包含具体项目、表达自然的三项并列没有问题
  • 互动诱导（“comment YES”、“tag 3 friends”、“double tap if”）
  • 断奏堆叠和为制造戏剧感而使用单词成行的句子（humanizer 指纹），或每一行都读起来机械、僵硬的段落

Example

示例

See 
references/examples.md
for worked before/after rewrites.
有关 caption 和 slide 的改写前后实例，参见 
references/examples.md
。

Files

文件

  • SKILL.md
    - this file (rewrite scrubber + audit-mode entry)
  • references/scrub-rules.md
    - V3 regex patterns by tier, density scoring, em dash cap, rhythm rules, forbidden insertions
  • references/examples.md
    - worked before/after rewrites for captions and slides
  • references/audit-checklist.md
    - the pre-publish checklist with thresholds
  • sub-skills/post-audit.md
    - pre-publish audit workflow (detection-only, no rewrite)
  • sub-skills/voice-profile.md
    - build/update the user's Voice & Brand Profile (
    --mode profile
    )
  • sub-skills/illustration.md
    - optional Pixfaro image workflow
  • SKILL.md
    - 本文件（改写清理器 + audit 模式入口）
  • references/scrub-rules.md
    - 按等级划分的 V3 regex 模式、密度评分、em dash 上限、节奏规则和禁止插入项
  • references/examples.md
    - caption 和 slide 的改写前后实例
  • references/audit-checklist.md
    - 包含阈值的发布前 checklist
  • sub-skills/post-audit.md
    - 发布前 audit 工作流（仅检测，不改写）
  • sub-skills/voice-profile.md
    - 建立或更新用户的 Voice & Brand Profile（
    --mode profile
    ）
  • sub-skills/illustration.md
    - 可选的 Pixfaro 图像工作流

Voice profile mode (
--mode profile
)

Voice profile 模式（
--mode profile
）

ig-humanizer --mode profile
builds or updates the user's Voice & Brand Profile at 
../../references/voice-profile.md
from 3-6 of their real Instagram posts pasted in (portable, no token) or, if a read token is set, from pulled activity. Once filled, every writing skill in this bundle drafts in the user's voice automatically. See 
sub-skills/voice-profile.md
. Triggers: "build my voice profile", "learn my voice".
ig-humanizer --mode profile
会根据用户粘贴的 3-6 条真实 Instagram 帖子，在 
../../references/voice-profile.md
中建立或更新用户的 Voice & Brand Profile（可移植，无需 token）；如果设置了 read token，也可以从拉取的活动记录中提取。填妥后，这个技能包中的每个 writing skill 都会自动使用用户的 voice 起草内容。参见 
sub-skills/voice-profile.md
。触发语：“build my voice profile”、“learn my voice”。

Related skills

相关 skills

  • ig-caption-writer
    - generates captions that already pass the humanizer
  • ig-carousel-planner
    - generates carousels that already pass the humanizer
  • ig-caption-writer
    - 生成已经通过 humanizer 检查的 captions
  • ig-carousel-planner
    - 生成已经通过 humanizer 检查的 carousels