de-aigc-skills
中英双语学术降 AIGC + 去水印 / bilingual academic de-AIGC and provenance skill for empirical papers in economics, management and the social sciences (Turnitin AI, GPTZero, 知网 AMLC, 万方, 维普).
Install / Use
npx skills add brycewang-stanford/Auto-Empirical-Research-Skills --skill 48-de-AIGC-skillsInstalls into whichever agent you are using.
SKILL.md
Installable skill definition
Quality Score
Category
AI & Machine LearningSupported Platforms
Our assessment of de-aigc-skills
de-aigc-skills scores 94/100 on our quality scale, 185th of 950 AI & Machine Learning skills we index (top 20%).
Its SKILL.md is 19 KB long, well organised into 16 sections with 3 code examples: a thorough specification that gives an agent plenty to work with.
With 4,360 GitHub stars, it is one of the more widely adopted skills in the catalogue.
Maintenance, license and trust
- The repository was last updated about 2 months ago, so de-aigc-skills is actively maintained.
- Our last check on 2026-11-16 found the source still online.
- No license is declared. By default that means all rights are reserved: you can read it, but reusing or redistributing it is not clearly permitted. Ask the author before building on it commercially.
- Its trust signals score 88/100, with 1 caution from licensing, adoption, age or documentation. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.
Safety scan
No issues foundOur scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful.
AI review by kimi-k2.7-code on 2026-09-29. Automated pattern scan on 2026-09-28. It catches known dangerous patterns, not every risk — read a skill before letting an agent act on it.
de-aigc-skills compared with similar skills
All 4 of these similar skills score higher than de-aigc-skills; compare them before choosing.
| Skill | Score | Stars | Updated | Format |
|---|---|---|---|---|
| de-aigc-skills (this skill)by brycewang-stanford | 94 | 4.4k | 54d ago | SKILL.md |
| claude-memby thedotmack | 100 | 117.6k | today | CLAUDE.md |
| Understand-Anythingby Egonex-AI | 100 | 92.8k | 1d ago | CLAUDE.md |
| headroomby headroomlabs-ai | 100 | 79.8k | today | CLAUDE.md |
| CowAgentby zhayujie | 100 | 48.0k | today | CLAUDE.md |
Frequently asked questions
- How do I install de-aigc-skills?
- Run
npx skills add brycewang-stanford/Auto-Empirical-Research-Skills --skill de-aigc-skills. The install tabs above show the steps for each supported agent. - Which AI agents does de-aigc-skills work with?
- It is written for Universal, as a SKILL.md file. Other agents that read the same format can often use it too.
- Is de-aigc-skills safe to use?
- Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful. It declares no license and scores 88/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
- Is de-aigc-skills still maintained?
- The repository was last updated about 2 months ago, so de-aigc-skills is actively maintained.
Skill content
View source on GitHubname: de-aigc-skills description: 中英双语学术降 AIGC / bilingual academic de-AIGC skill. Removes AI-generated writing signatures from empirical papers in economics, management, and the social sciences — in both English and Chinese. Covers Turnitin AI, GPTZero, Originality.ai on the English side and 知网 AMLC, 万方, 维普 on the Chinese side. Uses a six-step loop (intake → audit → claim-evidence check → differentiated rewrite → five-dimension self-score → cold-reader recheck) with two pattern libraries (22 English + 17 Chinese patterns), section-by-section strategies for empirical papers, and hard protections that keep every number, coefficient, and citation intact. Since 2026-08 it also carries a provenance layer (去水印): deterministic, CJK-safe cleaning of invisible-character carriers and file metadata (docx / png / jpg / svg / pdf, incl. C2PA), an honest account of Claude's statistical text watermark, and an author "ownership pass" — it never claims a text is watermark-free. triggers:
- de-AIGC
- 降 AIGC
- 降低 AIGC
- 去 AI 味
- 去 AI 痕迹
- humanize academic paper
- remove AI writing patterns
- AI detection academic
- 知网 AIGC 检测
- 万方 AIGC
- 维普检测
- Turnitin AI
- GPTZero
- humanize empirical paper
- 人工化重写
- 去水印
- 去 AI 水印
- Claude 水印
- remove Claude watermark
- AI watermark
- 零宽字符 / invisible Unicode allowed-tools:
- Bash(python3 scripts/provenance_scrub.py:*)
De-AIGC Skills · 中英双语学术降 AIGC
Restore the language distribution of a real researcher — in English or Chinese — for empirical papers in economics, management, and the social sciences. Not synonym swapping. Not sentence shuffling. Systematic reconstruction of the statistical signatures that mark a manuscript as AI-generated.
Scope 适用范围
Papers: empirical work in economics, management, finance, accounting, sociology, political science, education, public policy — anything built on data, identification, and regression tables. Theory papers and pure humanities essays are out of scope (most patterns still transfer, but the section strategies assume an empirical skeleton).
Languages:
- English manuscripts — targets Turnitin AI, GPTZero, Originality.ai, Copyleaks
- Chinese manuscripts — targets 知网 AMLC、万方、维普通达、Turnitin 中文版
- Mixed manuscripts — a Chinese paper with an English abstract, or a bilingual submission package: run each part through its own pattern library, then check cross-language consistency (the English abstract must not overclaim what the Chinese conclusion hedges, and vice versa)
Typical situations:
- Self-checking AIGC rate before journal submission (中文期刊投稿 / SSCI submission)
- Dissertations facing 知网 AMLC or a university's Turnitin AI screen
- Grant proposals, working papers, and reports drafted with AI assistance
- A human-written draft that detectors misclassify as AI (this happens often — formulaic academic prose looks like AI to n-gram detectors)
- Cleaning a submission package before (double-blind) review: invisible
characters picked up through copy-paste,
.docxauthor fields, C2PA manifests on model-generated figures — the provenance layer below
What does NOT work 无效做法
- ❌ Synonym replacement — detectors read n-gram distributions, not word lists. 把"关键"改成"核心"、"important" 改成 "crucial" 毫无作用。
- ❌ Sentence inversion — flipping "Because A, B" to "B, because A" leaves the syntactic template intact.
- ❌ Feeding the text to another AI for a "rewrite" — swaps one AI's signature for another's. Paraphraser tools are the fastest way to a higher AI score. Since 2026-08 it also re-embeds a statistical watermark: Claude's and Gemini's marks live in the word choices, so a model paraphrase is a fresh mark.
- ❌ Injecting typos or awkward grammar to "look human" — human experts do not write badly; graders and reviewers notice, and modern detectors are not fooled.
- ❌ Stripping invisible characters and calling the text "de-watermarked" — Claude's watermark is not made of characters (Anthropic: "there are no hidden characters"). Character hygiene is still necessary — carriers split tokens in n-gram scans and trip Turnitin's hidden-text flags — but it proves nothing about the watermark.
What works: targeted destruction of the structural signatures listed below, plus restoring the concrete, hedged, evidence-anchored voice of a real researcher.
Watermarks are not writing signatures 水印 ≠ 文风
Three different things are called an "AI watermark" in 2026. The skill treats
them differently, and every report says which layer was actually verified
(full policy and code-point tables: references/watermarks.md):
| Layer | What it is | Handled by |
|---|---|---|
| A · character carriers 隐藏字符 | zero-width characters, bidi controls, tag characters, stray variation selectors, exotic spaces, soft hyphens — not used by Claude, but common in copy-paste chains and third-party tools | scripts/provenance_scrub.py — deterministic and CJK-safe: keeps U+3000 indents, fullwidth punctuation, ideographic variation selectors in names, en dashes in ranges, R²; NFKC is not an option |
| B · statistical watermark 统计水印 | Claude (models launched ≥ 2026-08-02) and Gemini bias which words are sampled; nothing is added to the string; survives copy, translation and light edits; no public detector | the ownership pass in Step 3 — the author writes the high-signal sentences; the agent's own rewrite is model output and does not count |
| C · container metadata 容器元数据 | C2PA manifests on model-generated PNG / JPG / SVG, .docx docProps / comments / people.xml, PDF Info + XMP | provenance_scrub.py clean (docx / png / jpg / svg, lossless); report plus exiftool / qpdf commands for PDF; best of all, regenerate figures from the replication package |
python3 scripts/provenance_scrub.py inspect main.docx figures/*.png # report only; exit 1 if anything is found
python3 scripts/provenance_scrub.py clean main.docx --lang zh # writes main.clean.docx, docProps blanked
python3 scripts/provenance_scrub.py clean draft.md --diff # carriers shown as ⟨U+200B⟩ in the diff
python3 scripts/provenance_scrub.py self-test # 39 keep / strip fixtures
Every report ends with B: unknown. Nobody outside Anthropic can verify
that a statistical watermark is gone, and this skill will not claim it.
Structural signatures 结构性特征
The two languages fail differently. English LLM output leans on inflated significance and participle padding; Chinese LLM output leans on four-character formulas and connective scaffolding. The one signature they share — and the single highest-impact fix in either language — is uniform sentence rhythm.
English AI text (details and fixes: references/patterns-en.md, EN01–EN22):
- Inflated significance — "pivotal", "underscores the importance", "paves the way"
- Superficial "-ing" tails — "…, highlighting the need for further research"
- Formulaic scaffolding — "In recent years…", Moreover/Furthermore chains, rule-of-three lists, "not only… but also"
- Overclaiming verbs with underpowered evidence — "proves", "demonstrates", "establishes" hanging off an observational coefficient
- Low sentence-length variance — nearly all sentences 20–30 words
Chinese AI text (details and fixes: references/patterns-zh.md, ZH01–ZH17):
- 四字套话密度过高 — 每 200 字 3+ 个"综上所述/毋庸置疑/显而易见"
- 虚词与关联词冗余 — "此外/因此/而且/在此基础上"机械堆叠
- 主语回避 — 满篇"本文认为/相关研究表明",没有具体的研究者和文献
- 句长方差低 — 句子集中在 20–35 字,缺乏节奏跳跃
- 结论绝对化 — "充分证明了/必然导致/毫无疑问"
The six-step loop 六步闭环
0. 定位路由 1. 审计扫描 2. 主张-证据核对
Intake → Audit → Claim–evidence
│
5. 冷读复查 4. 五维自评 3. 差异化改写
Recheck ← Self-score ← Rewrite
Provenance layer, threaded through the loop: inspect before Step 1 (hidden
characters split tokens and corrupt the pattern scan) · ownership pass inside
Step 3 · package clean at Step 5.
Step 0 · Intake & routing 定位路由
Before touching the text:
- Detect language(s) — route each part to the right pattern library
(
patterns-en.md/patterns-zh.md). For mixed manuscripts, note which sections are which. - Map sections — abstract, intro, literature, data, empirical strategy,
results, mechanisms, robustness, discussion, conclusion. Rewrite intensity
differs sharply by section (
references/sections.md). - Identify the venue — a Chinese CSSCI journal, an SSCI field journal, and a dissertation committee have different tolerances for first-person voice and hedging. Ask the user if unclear.
- Ask for a voice sample (optional but powerful) — if the author has earlier human-written papers or paragraphs, match their sentence rhythm, connective habits, and hedging placement instead of a generic "human" style.
- Scan for provenance marks —
python3 scripts/provenance_scrub.py inspect <files>on the draft and any figures. Clean Layer A carriers before the audit; note docx / figure metadata for Step 5. Ask whether the draft (or any figure) came out of a model launched after 2026-08-02 — if so, the ownership pass in Step 3 is mandatory for the high-signal sections, not optional.
Step 1 · Audit scan 审计扫描
Scan the full text against both pattern libraries and output a structured audit report — do not edit anything yet. The author must see the whole picture first.
## AI-signature audit / AI 痕迹审计
| ¶ | Excerpt 原文片段 | Rule 规则 | Severity 严重度 |
|---|-----------------|----------|----------------|
| 2 | "毋庸置疑,数字化转型…" | ZH01 四字套话 | 🔴 |
| 5 | "…, underscoring the importance of digital…" | EN02 -ing tail | 🔴 |
| 7 | "This proves that the reform caused…" | EN10 overclaiming verb | 🔴 |
Include a summary line: total hits per severity, the 3 worst sections, and the
estimated rewrite depth (light polish / section rewrites / full-pass rewrite) —
and attach the provenance-mark report from Step 0 (Layer A counts by code point,
docx / figure metadata found, B: unknown).
Step 2 · Claim–evidence audit 主张-证据核对
Empirical papers live or die on the match between verbs and evidence strength. This step is what makes de-AIGC for empirical work different from generic humanizing:
- Every causal or quantitative claim must anchor to a number, table, figure, or citation. "显著提升企业绩效" → which table, which column, which coefficient?
- Verb ↔ design match:
- Clean identification (RCT, sharp RD, well-defended DiD) → direct statements are fine: "the reform reduced entry by 12%"
- Observational / correlational → "is associated with", "与…相关"
- Suggestive / mechanism evidence → "is consistent with", "为…提供了证据", "这与…的解释一致"
- Flag every "prove/demonstrate/establish/充分证明/必然导致" whose design cannot carry that weight — and every unsupported "significant/显著" with no test statistic.
- Never resolve a mismatch by inventing evidence. If a claim has no anchor, flag it for the author; weakening the verb is the default fix.
Step 3 · Differentiated rewrite 差异化改写
Work through the audit list, section by section, using the per-section strategies
in references/sections.md. Priorities, in order of impact:
- Break the rhythm ceiling first 先砸句长方差 — this is the single highest-leverage fix in both languages. Per ~200 words (or 200 字): at least one short sentence (≤8 words / ≤15 字) and one long sentence (≥40 words / ≥50 字). Short sentences open questions or land emphasis ("The data say otherwise." / "数据讲了另一个故事。"); long sentences carry the evidence.
- Concretize 具体化 — replace vague attributions and inflated adjectives with data, authors, years: "相关研究表明" → "Acemoglu and Restrepo (2020) estimate…"; "profound impact" → "raised TFP by 4.3% (t = 3.81)".
- De-scaffold 拆脚手架 — remove paragraph-initial connectives (Moreover/Furthermore/此外/因此) and ordina
Truncated for display — read the full file on GitHub.
Related Skills
claude-mem
117.6kPersistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + More
Understand-Anything
92.8kGraphs that teach > graphs that impress. Turn any code into an interactive knowledge graph you can explore, search, and ask questions about. Works with Claude Code, Codex, Cursor, Copilot, Gemini CLI, and more.
headroom
79.8kCompress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.
CowAgent
48.0kOpen-source personal AI assistant & Agent Harness. Plans tasks, runs tools and skills, self-evolves with memory and knowledge. Multi-agent, multi-model, multi-channel. Lightweight, extensible, one-line install.
Languages
Trust signals
From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.
