Skip to content

A local-first multi-agent app development platform, built on PenguinHarness

refactor(skills): split humanizer language cues into per-language files (#268)

Yaowei Zheng054cf4af2fe7
diff --git a/packages/skills/skills/humanizer/SKILL.md b/packages/skills/skills/humanizer/SKILL.md
index 209ed9de..e909a787 100644
--- a/packages/skills/skills/humanizer/SKILL.md
+++ b/packages/skills/skills/humanizer/SKILL.md
@@ -1,10 +1,10 @@
---
name: humanizer
-description: Rewrite or edit prose in any language so it reads like edited human writing in the register of books, newspapers and encyclopedias rather than default AI output. A small drafting core — vary every pattern, build density from anchored facts in whole grammar, cap the quotables, put a real writer with real material behind the text, let structure serve content, write each language from inside its idiom and typography, verify what you assert, and aim for the natural distribution of edited prose rather than a perfect scorecard — backed by a three-layer tell catalog, six-language cues and a five-round measured case study shipped as reference files for the diagnostic census.
+description: Rewrite or edit prose in any language so it reads like edited human writing in the register of books, newspapers and encyclopedias rather than default AI output. A small drafting core — vary every pattern, build density from anchored facts in whole grammar, cap the quotables, put a real writer with real material behind the text, let structure serve content, write each language from inside its idiom and typography, verify what you assert, and aim for the natural distribution of edited prose rather than a perfect scorecard — backed by a three-layer tell catalog, per-language cue files for six languages and a five-round measured case study shipped as reference files for the diagnostic census.
short_description: Strip AI-writing tells from prose in any language.
short_description_zh: 去除文本的 AI 味,把行文改成书籍、报纸、百科式的风格,任何语言通用。
preinstall: false
-version: 6
+version: 7
updated: 2026-08-11T00:00:00Z
---
@@ -23,7 +23,7 @@ If the invocation carries no text and no assignment, ask for one: the draft to e
3. **Cap the quotables.** One or two turned phrases can carry a piece, counting every shape: chiasmus, mirrored re-description, balanced antithesis, aphoristic kickers. State the thesis once and develop it with new material or cut the echo. Most paragraphs end flat, and an argument survives a dud.
4. **A real writer, with the material.** Opinion owns its anecdotes and risks a judgment of its own; reportage has stood somewhere; reference registers keep the writer invisible without chaperoning the reader. Claims sized to the evidence, honest hedges kept, references anchored (who, where, when), and one non-obvious source beats a second canonical one.
5. **Structure serves content.** Open on ground, not on a cold verdict and not on an era; paragraphs develop one idea across several sentences; headings and bullets appear only where content is genuinely enumerable; the piece ends where the information ends.
-6. **Write from inside the language.** Native idiom — a sentence that back-translates cleanly into another language was composed there, so recompose it — native punctuation at native frequency (a dash where the register expects one beats zero), and the venue's typography held consistently. Per-language budgets and surface forms: [`reference/language-cues.md`](reference/language-cues.md).
+6. **Write from inside the language.** Native idiom — a sentence that back-translates cleanly into another language was composed there, so recompose it — native punctuation at native frequency (a dash where the register expects one beats zero), and the venue's typography held consistently. Per-language budgets and surface forms: [`reference/language-cues.md`](reference/language-cues.md), which indexes one `reference/<lang>-cues.md` file per language.
7. **True, verified, calibrated.** Never invent specifics. Re-derive every mechanism example from the stated mechanism, split fused attributions, and verify or delete every superlative. One fluent falsehood outweighs any amount of style.
Over all seven: **aim for the natural distribution of edited prose, not a perfect scorecard.** Every rule above, overdriven, mints a new tell — the case study documents three such artifacts (banned openers became cold-open verdicts; scrubbed punctuation became semicolon inflation and rationed warmth; density became telegraph compression). Keep some looseness, an aside, an unresolved edge.
@@ -41,4 +41,4 @@ Before, tells marked: "In today's rapidly evolving AI landscape [era opener], sk
After: "A skill is a document that tells an AI agent how to perform one kind of task. The agent keeps only the document's one-line summary in memory and reads the full text when a matching task arrives, so hundreds of skills can be installed at negligible cost. The format is plain Markdown with a short metadata header, which means anyone who can write instructions can write a skill."
-The full tell catalog is in [`reference/tells.md`](reference/tells.md), per-language surface forms in [`reference/language-cues.md`](reference/language-cues.md), and the measured record behind all of it in [`reference/case-study.md`](reference/case-study.md).
+The full tell catalog is in [`reference/tells.md`](reference/tells.md), per-language surface forms in the `reference/<lang>-cues.md` files indexed by [`reference/language-cues.md`](reference/language-cues.md), and the measured record behind all of it in [`reference/case-study.md`](reference/case-study.md).
diff --git a/packages/skills/skills/humanizer/reference/de-cues.md b/packages/skills/skills/humanizer/reference/de-cues.md
new file mode 100644
index 00000000..fe39e302
--- /dev/null
+++ b/packages/skills/skills/humanizer/reference/de-cues.md
@@ -0,0 +1,18 @@
+# German (de)
+
+Default-AI markers:
+
+- „nicht nur …, sondern auch …" and the antithetic closer „weniger X als Y" as default endings.
+- „Doch"-pivots heading every paragraph; „sowohl … als auch" accumulation.
+- Nominalstil cascades („die Gewährleistung der Aufrechterhaltung der Bargeldversorgung").
+- „spielt eine zentrale Rolle", „In Zeiten von …", „Es bleibt abzuwarten, ob …", „Insgesamt zeigt sich ein differenziertes Bild."
+- Dreierreihung (schnell, sicher und bequem) beyond natural enumeration.
+- Vague Expertenattribution („Experten zufolge") where „gilt vielen als" or a named source belongs.
+- Gedankenstrich-Aphoristik („Bargeld ist geprägte Freiheit – bis heute").
+- Synonym rotation to dodge repetition („begleichen" for two euros because „bezahlen" was used already).
+
+Scrubbed-text markers: zero Gedankenstriche in a genre that loves them; quoteless journalism — no named person, no direct quote, respondents existing only as aggregate; one-move-per-paragraph architecture too clean to be written.
+
+Native, not a tell: verbal style with occasional compounds; one antithetic closer; Doppelpunkt setups.
+
+Print norms: „deutsche Anführungszeichen" or »Chevrons« per house; the Gedankenstrich is the en dash with spaces ( – ), not the em dash.
diff --git a/packages/skills/skills/humanizer/reference/en-cues.md b/packages/skills/skills/humanizer/reference/en-cues.md
new file mode 100644
index 00000000..7a06d859
--- /dev/null
+++ b/packages/skills/skills/humanizer/reference/en-cues.md
@@ -0,0 +1,22 @@
+# English (en)
+
+Default-AI markers:
+
+- "not just X, but Y" and every correction-shaped frame, including the strawman-swerve ("was not trying to X; he wanted Y" against a position nobody holds).
+- Epigram saturation: every paragraph ends quotable; chiasmus and mirrored re-description ("engages with detail / the detail it engages with"); "X dressed up as Y"; zero duds anywhere.
+- Rule of three at every scale, including the "vivid concrete triad" with an expanding final item ("the sick week, the flaky test suite, the dependency upgrade that breaks the build").
+- Maxim anaphora ("Shipping beats planning. Feedback beats instinct. Done beats perfect.").
+- Em-dash saturation — or, after a scrub, em-dash extermination with semicolon inflation.
+- Scope-widener transitions ("The implications extend far beyond…"), anadiplosis echo ledes ("Piece by piece was the whole problem.").
+- Bookend kickers ("What began as X now Y") and "That, in the end, is…: not A, but B" ring closers.
+- Testament-speak and pet abstractions: stands as a testament, tapestry, landscape, inflection point, connective tissue.
+- Participial analysis tails ("…, highlighting the challenges facing the sector").
+- Vague authority ("experts agree"), hedge inflation ("arguably one of the most"), era openers ("In today's fast-paced world").
+- Elegant-variation epithet cycling (the device… the ancient computer… the bronze marvel).
+- Mechanical concession pivot ("To be sure, challenges remain. But…") and "remains to be seen" closers.
+
+Scrubbed-text markers: dash and parenthesis blackout in registers that breathe through both; comma splices doing dash work ("did not improve the old system, they replaced it"); first-person amputation in opinion genres; ownerless anecdotes ("The feature," no company, no year); fragments For Effect.
+
+Native, not a tell: the occasional triad, one earned antithesis, semicolons in essay register — at frequencies a human editor would pass.
+
+Print norms: pick one style guide and hold it; "26 April 1956" alongside "aluminum" is a pipeline fingerprint. Parentheses for abbreviations: "(TEU)", not ", TEU."
diff --git a/packages/skills/skills/humanizer/reference/es-cues.md b/packages/skills/skills/humanizer/reference/es-cues.md
new file mode 100644
index 00000000..a918ed08
--- /dev/null
+++ b/packages/skills/skills/humanizer/reference/es-cues.md
@@ -0,0 +1,18 @@
+# Spanish (es)
+
+Default-AI markers:
+
+- "No solo… sino (también)…" scaffolds; identity-elevation closers ("Más que una escritora, fue la voz de toda una generación").
+- Obituary-speak: legado, huella imborrable, figura fundamental, "un antes y un después".
+- Filler of emphasis: cabe destacar, es importante señalar, sin duda alguna.
+- Abstract triads (talento, esfuerzo y perseverancia); "a lo largo de" as universal time connector.
+- Harvest clichés: cosechar éxitos/elogios/reconocimientos.
+- Gerund-tail calques of English -ing glosses ("…, consolidando su prestigio internacional").
+- Era openers ("En un mundo cada vez más digital…"); "verdadero/a" inflation.
+- Anglicisms of register: "eventualmente" for finalmente, "asumir" for suponer.
+
+Scrubbed-text markers: no raya (—) anywhere across literary-register prose (Spanish expects an occasional inciso entre rayas); evaluative reticence — a profile or review that outsources every aesthetic judgment to quoted sources and never risks one clause of its own.
+
+Native, not a tell: the triadic metonymy of the semblanza tradition ("cuatro décadas de aulas, consulados y cuadernos"); ring composition around a governing epithet; personification of events ("El premio la encontró de viaje").
+
+Print norms: comillas angulares « » in print; the raya for incisos and dialogue; question and exclamation marks opened (¿ ¡).
diff --git a/packages/skills/skills/humanizer/reference/fr-cues.md b/packages/skills/skills/humanizer/reference/fr-cues.md
new file mode 100644
index 00000000..fe3fa8a7
--- /dev/null
+++ b/packages/skills/skills/humanizer/reference/fr-cues.md
@@ -0,0 +1,18 @@
+# French (fr)
+
+Default-AI markers:
+
+- « Il ne s'agit pas de X, mais de Y » and the aphoristic balance « moins… que… » (« Tout se joue moins en vitrine que dans l'atelier »).
+- The sensory triad after a colon (« l'odeur, le geste, la conversation »).
+- Antithesis as the only engine: four paragraphs, four load-bearing reversals.
+- The dramatic verbless reveal (« Derrière la célébration, une inquiétude : … ») and self-certifying hedges (« bien documentée » with no document).
+- « Force est de constater que », « véritable » inflation, « au cœur de », « enjeu majeur ».
+- Analysis verbs as filler: témoigne de, illustre, incarne.
+- « À l'heure où / dans un monde où » openers; « autant de » recaps; the « n'a pas dit son dernier mot » kicker.
+- The panoramic millions-lede (« Chaque matin, des millions de gens répètent le même geste… »).
+
+Scrubbed-text markers: comma-apposition where French print wants the colon (« quatre ingrédients, farine, eau, sel »); unnamed collective attributions delivering perfectly turned ironies (« Les boulangers, eux, font remarquer que… » with no name and no apron).
+
+Native, not a tell: one antithesis done well; colon reveals; the dislocation « Les boulangers, eux, » — French essay prose is legitimately more rhetorical than English; judge frequency, not presence.
+
+Print norms: « guillemets » with non-breaking spaces; passé simple belongs to essay and narration, not reportage; reportage requires a scene and a named person.
diff --git a/packages/skills/skills/humanizer/reference/ja-cues.md b/packages/skills/skills/humanizer/reference/ja-cues.md
new file mode 100644
index 00000000..0375e5d7
--- /dev/null
+++ b/packages/skills/skills/humanizer/reference/ja-cues.md
@@ -0,0 +1,22 @@
+# Japanese (ja)
+
+Default-AI markers:
+
+- Genre-blind です・ます where print demands だ・である (新書, コラム, 科学読み物); register bleed inside 常体 (a stray でしょう in a である paragraph).
+- Verdict formulas closing paragraphs: 〜と言えるでしょう、まさに〜と言えます.
+- 「〜することができます」 potential bloat for plain potentials (感じ取れる).
+- Recap scaffolding: このように、まとめると、いかがでしたか.
+- Significance announcements: 重要なのは、注目すべきは、鍵となるのは.
+- Vagueness padding: さまざまな、多くの、〜など wrapped around a triple list.
+- Rhetorical reader-contact closers: 〜ではないでしょうか.
+- Contrast frames as default argument: AだけでなくBも、AではなくB.
+- のです/のである overuse: explanatory tone on every third sentence.
+- カタカナ語 translationese lean: メカニズム、プロセス、ナビゲーション where 仕組み・過程・経路 belong; calqued possessives (彼らの旅).
+- Paragraph-initial connective metronome: また/さらに/一方で/加えて heading every paragraph.
+- Thesis-forecast outlines: a tidy binary announced in the lede (外と内の両方にある), then executed capsule by capsule.
+
+Scrubbed-text markers: machine-gun 体言止め and clipped drama beats (頼りは星。そして磁場。) installed where clichés were removed; hermetic impersonality (no gloss on first-use technical terms, no researcher given an affiliation, no aside); priority-cleft series as the sole engine (〜を示したのは…だった。〜を作ったのは…で、〜を突き止めたのは…である。 three in a row).
+
+Native, not a tell: です・ます in service and consumer registers; one 〜だったのである for emphasis; 体言止め in headlines and captions.
+
+Print norms: no half-width spaces between ASCII and CJK (「1960年代」, not 「1960 年代」); colons in running text are pipeline typography; species names in katakana.
diff --git a/packages/skills/skills/humanizer/reference/language-cues.md b/packages/skills/skills/humanizer/reference/language-cues.md
index a44acc98..14d53f95 100644
--- a/packages/skills/skills/humanizer/reference/language-cues.md
+++ b/packages/skills/skills/humanizer/reference/language-cues.md
@@ -1,145 +1,17 @@
# Language cues
-The tells in SKILL.md are structural and hold across languages; what varies is the surface. This file records, per language, how the universal tells surface locally, which over-correction signs to watch for, which native rhetoric is NOT a tell, and the print norms that betray pipeline typography. The lists were harvested from independent editorial reviews in the round-three field tests (see `case-study.md`) and are ranked roughly by how strongly each marker signals machine writing. For a language not listed, map the structural pattern and ask what its editors would flag.
+The tells in SKILL.md are structural and hold across languages; what varies is the surface. The sibling `<lang>-cues.md` files record, one per language, how the universal tells surface locally, which over-correction signs to watch for, which native rhetoric is NOT a tell, and the print norms that betray pipeline typography. The lists were harvested from independent editorial reviews in the round-three field tests (see `case-study.md`) and are ranked roughly by how strongly each marker signals machine writing. For a language not listed, map the structural pattern and ask what its editors would flag.
-Two rules frame everything below:
+Two rules frame every language file:
- **Budgets are language-relative.** A dash count that is right for English print is wrong for Spanish, which expects an occasional inciso entre rayas; French leans on colon reveals; German feuilleton uses the Gedankenstrich. Zero everywhere is not neutral, it is the scrub showing.
- **Native rhetoric in native doses is not a tell.** French antithesis, Spanish triadic metonymy in a semblanza, Japanese です・ます in service registers, Chinese 四字格 in titles: these belong to their traditions. The tell is frequency and function, not existence.
-## Chinese (zh)
+Per-language files:
-Default-AI markers:
-
-- 「不是……而是……」及其亲属(「不需要 X,只需要 Y」「这不仅是……更是……」)翻案句。
-- 首尾升华:宏大开场(在……的今天)与拔高结尾(未来已来;……的故事仍在继续)。
-- 三连排比与四字格堆叠(一种态度、一种选择、一种生活方式;源远流长、波澜壮阔)。
-- 显性路标(首先/其次/最后、总而言之、值得注意的是)与设问自答开头。
-- 「这意味着/这标志着」自我注解句:先陈述,再补一句解释其意义。
-- 象征膨胀动词:见证了、承载着、彰显了、是……的缩影。
-- 机构腔与黑话:赋能、打造、助力、闭环、抓手、擦亮名片;公文动词 进行/加以。
-- 破折号与冒号定义腔的移植(「手冲的本质:一场与时间的对话」)。
-- 万能安全垫结尾(具体效果因人而异,建议根据自身情况调整)。
-- 伪 hedge(某种意义上、一定程度上、或许正是)与幽灵人群(比多数人预想的快——谁预想?)。
-- 欧化搭配与翻译腔:作为……的身份就此终结、死档案、密集的簇;欧化长定语。
-- 信息复读:换句话说/也就是说 引出的同义重述。
-
-Scrubbed-text markers (over-correction):
-
-- 冒号、破折号、引语、史料味全部清零,逗号串承担一切(七个分句一逗到底)。
-- 口语点缀按段配给(每段恰好一个"接地气"词),温度像用滴管加的。
-- 段段短句收尾的"留白"节拍器(人写的段落有时就平平地结束)。
-- 判断句系列:连续段落都用「……的是X」承重(把流程定下来的是创始团队。真正把成本压下来的是第二代产线。),同一句型当引擎。
-- 电报体编年:连续分句各带一个年份(1997年……。1998年……;2000年……,2001年……),叙事被压成 changelog;改用相对时间(三年后、世纪末)并让重要事件独占一句。
-- 分号串联叙事事件:中文散文里分号几乎只属于正式的并列列举,拿来串故事就是压缩痕迹。
-- 无主语压缩:「初代销量不错」——谁的初代?首次提及必须带全称,指代不能靠上一段撑着。
-- 等重短句链:四个同构主谓短句一口气连排,虽无重复虚词,读感仍是排比;把其中一件事降为从句,节拍就破了。
-- 紧缩四字支点:「接口一变」「需求一改」式的四字条件枢纽,后面挂「就得/都得跟着……」的尾巴;展开成「一旦……」或直接拆句。
-- 「就」字链:就/就得/都得/一……就…… 承担全篇的因果衔接,单个地道、高频机械;每段数一数「就」,用拆句、从句或自带结果的动词分担。
-
-Native, not a tell: 四字格 in titles and set phrases at natural density; topic-comment comma chains of moderate length; 顿号 enumerations of real items.
-
-Print norms: mainland print typically sets Arabic numerals tight (「公元605年」); spacing ASCII from CJK (「605 年」) is a tech-blog convention — follow the venue.
-
-## English (en)
-
-Default-AI markers:
-
-- "not just X, but Y" and every correction-shaped frame, including the strawman-swerve ("was not trying to X; he wanted Y" against a position nobody holds).
-- Epigram saturation: every paragraph ends quotable; chiasmus and mirrored re-description ("engages with detail / the detail it engages with"); "X dressed up as Y"; zero duds anywhere.
-- Rule of three at every scale, including the "vivid concrete triad" with an expanding final item ("the sick week, the flaky test suite, the dependency upgrade that breaks the build").
-- Maxim anaphora ("Shipping beats planning. Feedback beats instinct. Done beats perfect.").
-- Em-dash saturation — or, after a scrub, em-dash extermination with semicolon inflation.
-- Scope-widener transitions ("The implications extend far beyond…"), anadiplosis echo ledes ("Piece by piece was the whole problem.").
-- Bookend kickers ("What began as X now Y") and "That, in the end, is…: not A, but B" ring closers.
-- Testament-speak and pet abstractions: stands as a testament, tapestry, landscape, inflection point, connective tissue.
-- Participial analysis tails ("…, highlighting the challenges facing the sector").
-- Vague authority ("experts agree"), hedge inflation ("arguably one of the most"), era openers ("In today's fast-paced world").
-- Elegant-variation epithet cycling (the device… the ancient computer… the bronze marvel).
-- Mechanical concession pivot ("To be sure, challenges remain. But…") and "remains to be seen" closers.
-
-Scrubbed-text markers: dash and parenthesis blackout in registers that breathe through both; comma splices doing dash work ("did not improve the old system, they replaced it"); first-person amputation in opinion genres; ownerless anecdotes ("The feature," no company, no year); fragments For Effect.
-
-Native, not a tell: the occasional triad, one earned antithesis, semicolons in essay register — at frequencies a human editor would pass.
-
-Print norms: pick one style guide and hold it; "26 April 1956" alongside "aluminum" is a pipeline fingerprint. Parentheses for abbreviations: "(TEU)", not ", TEU."
-
-## Japanese (ja)
-
-Default-AI markers:
-
-- Genre-blind です・ます where print demands だ・である (新書, コラム, 科学読み物); register bleed inside 常体 (a stray でしょう in a である paragraph).
-- Verdict formulas closing paragraphs: 〜と言えるでしょう、まさに〜と言えます.
-- 「〜することができます」 potential bloat for plain potentials (感じ取れる).
-- Recap scaffolding: このように、まとめると、いかがでしたか.
-- Significance announcements: 重要なのは、注目すべきは、鍵となるのは.
-- Vagueness padding: さまざまな、多くの、〜など wrapped around a triple list.
-- Rhetorical reader-contact closers: 〜ではないでしょうか.
-- Contrast frames as default argument: AだけでなくBも、AではなくB.
-- のです/のである overuse: explanatory tone on every third sentence.
-- カタカナ語 translationese lean: メカニズム、プロセス、ナビゲーション where 仕組み・過程・経路 belong; calqued possessives (彼らの旅).
-- Paragraph-initial connective metronome: また/さらに/一方で/加えて heading every paragraph.
-- Thesis-forecast outlines: a tidy binary announced in the lede (外と内の両方にある), then executed capsule by capsule.
-
-Scrubbed-text markers: machine-gun 体言止め and clipped drama beats (頼りは星。そして磁場。) installed where clichés were removed; hermetic impersonality (no gloss on first-use technical terms, no researcher given an affiliation, no aside); priority-cleft series as the sole engine (〜を示したのは…だった。〜を作ったのは…で、〜を突き止めたのは…である。 three in a row).
-
-Native, not a tell: です・ます in service and consumer registers; one 〜だったのである for emphasis; 体言止め in headlines and captions.
-
-Print norms: no half-width spaces between ASCII and CJK (「1960年代」, not 「1960 年代」); colons in running text are pipeline typography; species names in katakana.
-
-## French (fr)
-
-Default-AI markers:
-
-- « Il ne s'agit pas de X, mais de Y » and the aphoristic balance « moins… que… » (« Tout se joue moins en vitrine que dans l'atelier »).
-- The sensory triad after a colon (« l'odeur, le geste, la conversation »).
-- Antithesis as the only engine: four paragraphs, four load-bearing reversals.
-- The dramatic verbless reveal (« Derrière la célébration, une inquiétude : … ») and self-certifying hedges (« bien documentée » with no document).
-- « Force est de constater que », « véritable » inflation, « au cœur de », « enjeu majeur ».
-- Analysis verbs as filler: témoigne de, illustre, incarne.
-- « À l'heure où / dans un monde où » openers; « autant de » recaps; the « n'a pas dit son dernier mot » kicker.
-- The panoramic millions-lede (« Chaque matin, des millions de gens répètent le même geste… »).
-
-Scrubbed-text markers: comma-apposition where French print wants the colon (« quatre ingrédients, farine, eau, sel »); unnamed collective attributions delivering perfectly turned ironies (« Les boulangers, eux, font remarquer que… » with no name and no apron).
-
-Native, not a tell: one antithesis done well; colon reveals; the dislocation « Les boulangers, eux, » — French essay prose is legitimately more rhetorical than English; judge frequency, not presence.
-
-Print norms: « guillemets » with non-breaking spaces; passé simple belongs to essay and narration, not reportage; reportage requires a scene and a named person.
-
-## German (de)
-
-Default-AI markers:
-
-- „nicht nur …, sondern auch …" and the antithetic closer „weniger X als Y" as default endings.
-- „Doch"-pivots heading every paragraph; „sowohl … als auch" accumulation.
-- Nominalstil cascades („die Gewährleistung der Aufrechterhaltung der Bargeldversorgung").
-- „spielt eine zentrale Rolle", „In Zeiten von …", „Es bleibt abzuwarten, ob …", „Insgesamt zeigt sich ein differenziertes Bild."
-- Dreierreihung (schnell, sicher und bequem) beyond natural enumeration.
-- Vague Expertenattribution („Experten zufolge") where „gilt vielen als" or a named source belongs.
-- Gedankenstrich-Aphoristik („Bargeld ist geprägte Freiheit – bis heute").
-- Synonym rotation to dodge repetition („begleichen" for two euros because „bezahlen" was used already).
-
-Scrubbed-text markers: zero Gedankenstriche in a genre that loves them; quoteless journalism — no named person, no direct quote, respondents existing only as aggregate; one-move-per-paragraph architecture too clean to be written.
-
-Native, not a tell: verbal style with occasional compounds; one antithetic closer; Doppelpunkt setups.
-
-Print norms: „deutsche Anführungszeichen" or »Chevrons« per house; the Gedankenstrich is the en dash with spaces ( – ), not the em dash.
-
-## Spanish (es)
-
-Default-AI markers:
-
-- "No solo… sino (también)…" scaffolds; identity-elevation closers ("Más que una escritora, fue la voz de toda una generación").
-- Obituary-speak: legado, huella imborrable, figura fundamental, "un antes y un después".
-- Filler of emphasis: cabe destacar, es importante señalar, sin duda alguna.
-- Abstract triads (talento, esfuerzo y perseverancia); "a lo largo de" as universal time connector.
-- Harvest clichés: cosechar éxitos/elogios/reconocimientos.
-- Gerund-tail calques of English -ing glosses ("…, consolidando su prestigio internacional").
-- Era openers ("En un mundo cada vez más digital…"); "verdadero/a" inflation.
-- Anglicisms of register: "eventualmente" for finalmente, "asumir" for suponer.
-
-Scrubbed-text markers: no raya (—) anywhere across literary-register prose (Spanish expects an occasional inciso entre rayas); evaluative reticence — a profile or review that outsources every aesthetic judgment to quoted sources and never risks one clause of its own.
-
-Native, not a tell: the triadic metonymy of the semblanza tradition ("cuatro décadas de aulas, consulados y cuadernos"); ring composition around a governing epithet; personification of events ("El premio la encontró de viaje").
-
-Print norms: comillas angulares « » in print; the raya for incisos and dialogue; question and exclamation marks opened (¿ ¡).
+- [Chinese (zh)](zh-cues.md)
+- [English (en)](en-cues.md)
+- [Japanese (ja)](ja-cues.md)
+- [French (fr)](fr-cues.md)
+- [German (de)](de-cues.md)
+- [Spanish (es)](es-cues.md)
diff --git a/packages/skills/skills/humanizer/reference/zh-cues.md b/packages/skills/skills/humanizer/reference/zh-cues.md
new file mode 100644
index 00000000..38d9903c
--- /dev/null
+++ b/packages/skills/skills/humanizer/reference/zh-cues.md
@@ -0,0 +1,33 @@
+# Chinese (zh)
+
+Default-AI markers:
+
+- 「不是……而是……」及其亲属(「不需要 X,只需要 Y」「这不仅是……更是……」)翻案句。
+- 首尾升华:宏大开场(在……的今天)与拔高结尾(未来已来;……的故事仍在继续)。
+- 三连排比与四字格堆叠(一种态度、一种选择、一种生活方式;源远流长、波澜壮阔)。
+- 显性路标(首先/其次/最后、总而言之、值得注意的是)与设问自答开头。
+- 「这意味着/这标志着」自我注解句:先陈述,再补一句解释其意义。
+- 象征膨胀动词:见证了、承载着、彰显了、是……的缩影。
+- 机构腔与黑话:赋能、打造、助力、闭环、抓手、擦亮名片;公文动词 进行/加以。
+- 破折号与冒号定义腔的移植(「手冲的本质:一场与时间的对话」)。
+- 万能安全垫结尾(具体效果因人而异,建议根据自身情况调整)。
+- 伪 hedge(某种意义上、一定程度上、或许正是)与幽灵人群(比多数人预想的快——谁预想?)。
+- 欧化搭配与翻译腔:作为……的身份就此终结、死档案、密集的簇;欧化长定语。
+- 信息复读:换句话说/也就是说 引出的同义重述。
+
+Scrubbed-text markers (over-correction):
+
+- 冒号、破折号、引语、史料味全部清零,逗号串承担一切(七个分句一逗到底)。
+- 口语点缀按段配给(每段恰好一个"接地气"词),温度像用滴管加的。
+- 段段短句收尾的"留白"节拍器(人写的段落有时就平平地结束)。
+- 判断句系列:连续段落都用「……的是X」承重(把流程定下来的是创始团队。真正把成本压下来的是第二代产线。),同一句型当引擎。
+- 电报体编年:连续分句各带一个年份(1997年……。1998年……;2000年……,2001年……),叙事被压成 changelog;改用相对时间(三年后、世纪末)并让重要事件独占一句。
+- 分号串联叙事事件:中文散文里分号几乎只属于正式的并列列举,拿来串故事就是压缩痕迹。
+- 无主语压缩:「初代销量不错」——谁的初代?首次提及必须带全称,指代不能靠上一段撑着。
+- 等重短句链:四个同构主谓短句一口气连排,虽无重复虚词,读感仍是排比;把其中一件事降为从句,节拍就破了。
+- 紧缩四字支点:「接口一变」「需求一改」式的四字条件枢纽,后面挂「就得/都得跟着……」的尾巴;展开成「一旦……」或直接拆句。
+- 「就」字链:就/就得/都得/一……就…… 承担全篇的因果衔接,单个地道、高频机械;每段数一数「就」,用拆句、从句或自带结果的动词分担。
+
+Native, not a tell: 四字格 in titles and set phrases at natural density; topic-comment comma chains of moderate length; 顿号 enumerations of real items.
+
+Print norms: mainland print typically sets Arabic numerals tight (「公元605年」); spacing ASCII from CJK (「605 年」) is a tech-blog convention — follow the venue.