A paper-writing pipeline for Claude Code went from 6,400 GitHub stars to 51,000 in five months
academic-research-skills runs 32 agents across ten stages with two integrity gates. Its builder says the checks caught 15 fabricated citations in one real paper.
QbitAI (量子位) reported on 17 May 2026 on academic-research-skills, an open source project built by Edward Cheng-I Wu (吴政宜) of Taiwan. The article is in Chinese, and the translation is ours.
| Skill | Agents | It covers |
|---|---|---|
| Deep Research (深度研究) | 13 | Literature review, PRISMA reviews, citation verification |
| Academic Paper (学术写作) | 12 | Drafting, bilingual abstracts, APA 7.0 and IEEE output |
| Academic Paper Reviewer (学术审稿) | 7 | Review with a score from 0 to 100 |
| Academic Pipeline (流程编排) | Ten stages, with integrity gates at stages 2.5 and 4.5 |
The builder recommends Claude Opus 4.7 on a Max plan. A full run of all ten stages uses more than 200,000 input tokens and 100,000 output tokens, and the article puts a 15,000-character paper at roughly 4 to 6 US dollars.
The article lists limits the builder states. Output is not reproducible byte for byte. A gate can be overridden by hand, and the override is logged. The README’s principle is “AI is your copilot, not the pilot” (AI是你的副驾驶，不是飞行员).
The article gives no acceptance record for papers produced with the tool, and no outside evaluation. The repository was created on 26 February 2026 and was last updated on 9 October.
Sources
- QbitAI qbitai.com
- GitHub, academic-research-skills github.com