skills/48-copaper-ai-chinese-de-aigc/SKILL.md
面向中文学术论文的降 AIGC 检测率 Skill。针对知网、万方、维普、Turnitin 中文版的检测机制,识别并消除中文大语言模型的 17 类结构化写作痕迹。采用"定位 → 诊断 → 改写 → 自评 → 复查"五步闭环工作流,分章节差异化策略(摘要/引言/文献综述/方法/结果/讨论/结论),保持学术严谨性前提下通过检测。
npx skillsauth add brycewang-stanford/Awesome-Agent-Skills-for-Empirical-Research chinese-de-aigcInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
中文学术实证论文的 AI 痕迹消除器。不是"改同义词",不是"打乱语序",而是系统性地重构中文 AI 文本的统计学特征,让它回归到真实研究者写作的语言分布上。
三个错误的做法(很多教程都在做但无效):
本 Skill 的正确路径:针对中文 AI 的五大结构性特征做定点破坏,而非字词层面的表面修改。
与英文 AI 不同,中文大模型的痕迹主要表现在:
本 Skill 针对上述 5 大特征,提供 17 类细分诊断规则(详见 references/patterns.md)。
┌─────────────┐ ┌─────────────┐ ┌─────────────┐
│ 1. 定位扫描 │ → │ 2. 诊断分类 │ → │ 3. 差异化改写 │
└─────────────┘ └─────────────┘ └─────────────┘
│
┌─────────────┐ ┌─────────────┐ │
│ 5. 二次复查 │ ← │ 4. 五维自评 │ ←────────┘
└─────────────┘ └─────────────┘
接收用户提交的文本,按 references/patterns.md 的 17 类规则做全文扫描,输出结构化问题清单:
## AI 痕迹定位报告
| 段落 | 原文片段 | 命中规则 | 严重度 |
|------|---------|---------|--------|
| ¶2 | "毋庸置疑,数字化转型..." | P01 四字套话 | 高 |
| ¶3 | "...此外,该研究还..." | P04 显性连接词 | 中 |
| ¶5 | "本文认为该机制充分证明了..." | P12 绝对化断言 | 高 |
⚠️ 不要此时就开始改写,先让用户/作者看到问题全貌。
按段落功能分类:
参考 references/academic-sections.md 的分章节策略表决定每段的改写力度。
针对 Step 1 清单里的每一条,按以下四条原则逐一修复:
⚠️ 禁止事项:
对改写后的文本做中文学术版 5 维评分(每维 1-10 分,详见 references/scoring.md):
| 维度 | 检查点 | 权重 | |------|--------|------| | 具体性 | 是否用具体数据/案例/作者替代了模糊表达 | 1.5× | | 节奏性 | 句长方差是否 ≥ 150(50 字长句 + 15 字短句混排) | 1.2× | | 谨慎性 | 绝对化断言是否已降级为条件化表述 | 1.3× | | 隐衔接 | 段落之间是否消除了显性关联词(此外/因此等) | 1.0× | | 研究者语气 | 是否出现"我们/本团队/我"等第一人称研究立场 | 1.0× |
加权总分 < 35 → 返回 Step 3 再改一轮。加权总分 ≥ 42 → 通过。
用"冷读者"视角重新审视全文,执行三项终审:
输出终稿 + 改动摘要(哪些段落改了多少、为什么改)。
用户可以用以下任意触发词调用:
请对这段文本降 AIGC 检测率把这篇论文改得不像 AI 写的走 chinese-de-aigc 五步闭环诊断这段文字的 AI 痕迹,给出修改建议references/patterns.md — 17 类中文 AI 痕迹模式库(每类含识别规则 + 典型样本)references/examples.md — 12 组原文/改写前后对比(覆盖实证论文七个主要章节)references/academic-sections.md — 按章节差异化的改写策略表(摘要/引言/文献综述/方法/结果/讨论/结论)references/scoring.md — 五维评分量表细则本 Skill 的目标是让人工写作和 AI 辅助写作的文本回归到真实研究者的语言分布,而不是"帮 AI 生成内容骗过检测"。
学术诚信优先于检测率。任何改写都不应触及研究结论、数据真实性、引用准确性。
development
Conduct rigorous thematic analysis (TA) of qualitative data following Braun and Clarke's (2006) six-phase framework. Use whenever the user mentions 'thematic analysis', 'TA', 'Braun and Clarke', 'qualitative coding', 'identifying themes', or asks for help analysing interviews, focus groups, open-ended survey responses, or transcripts to identify patterns. Also trigger for questions about inductive vs theoretical coding, semantic vs latent themes, essentialist vs constructionist epistemology, building a thematic map, or writing up a qualitative findings section. Covers all six phases, the four upfront analytic decisions, the 15-point quality checklist, and the five common pitfalls. Produces a Word document write-up and an annotated thematic map. Does NOT cover IPA, grounded theory, discourse analysis, conversation analysis, or narrative analysis — use a different method for those.
development
Guide users through writing a systematic literature review (SLR) following the PRISMA 2020 framework. Use this skill whenever the user mentions 'systematic review', 'systematic literature review', 'SLR', 'PRISMA', 'PRISMA 2020', 'PRISMA flow diagram', 'PRISMA checklist', or asks for help writing, structuring, or auditing a literature review that follows reporting guidelines. Also trigger when the user asks about inclusion/exclusion criteria for a review, search strategies for databases like Scopus/WoS/PubMed, study selection processes, risk of bias assessment, or narrative synthesis for a review paper. This skill covers the full PRISMA 2020 checklist (27 items), produces a Word document manuscript in strict journal article format, generates an annotated PRISMA flow diagram, and enforces APA 7th Edition referencing throughout. It does NOT cover meta-analysis or statistical pooling. By Chuah Kee Man.
testing
Performs placebo-in-time sensitivity analysis with hierarchical null model and optional Bayesian assurance. Use when checking model robustness, verifying lack of pre-intervention effects, or estimating study power.
data-ai
Fit, summarize, plot, and interpret a chosen CausalPy experiment. Use after the causal method has been selected, including when configuring PyMC/sklearn models and scale-aware custom priors.