A manuscript construction skill suite (GitHub: WUBING2023/PaperSpine, ~240 stars as of May 2026, created May 2026) that runs inside Codex ($paper-spine) and Claude Code (/paperspine). Its defining feature: it refuses to start writing until a confirmed_motivation.md exists. Every section, every claim, and every revision is traced back to that central motivation — producing a writing rationale matrix that documents not just what each section says but why it belongs in the paper at all.
Source: [Weibo post, May 2026] + github.com/WUBING2023/PaperSpine
What Problem It Solves
Most AI writing tools start from a prompt: “write me an introduction for a paper about X.” PaperSpine treats this as the wrong starting point. A paper isn’t a collection of sections — it’s an argument, and an argument has a spine. Without confirmed motivation, individual sections can be well-written while the paper as a whole fails to cohere.
The toolkit enforces a motivation-first discipline:
Confirm motivation → Research SOTA → Build evidence bank
→ Section blueprint → Writing rationale matrix
→ Draft section by section → Revision audit trail
→ Final output (LaTeX + .bib, optional DOCX, optional PDF, optional CN translation)
Every artifact in the pipeline answers to the central motivation. If a section can’t be justified against the motivation, the rationale matrix exposes it before the writing happens.
Two First-Class Workflows
Workflow 1: Rewrite Existing You have a manuscript — a draft, a rejected submission, a course report that needs journal quality. PaperSpine audits the existing structure against the motivation it extracts, identifies what to restructure or cut, and rebuilds with a rationale matrix that documents every change.
Workflow 2: Build From Materials You have notes, figures, data summaries, partial drafts, PDFs — but no finished manuscript. PaperSpine assembles the paper from raw inputs, confirming motivation first, then constructing the evidence bank before a word of the main text is written.
Both workflows produce the same artifact set at the end. The difference is the starting point.
The Central Artifact: Writing Rationale Matrix
The writing_rationale_matrix.md is PaperSpine’s signature output and the thing that distinguishes it from general-purpose writing assistants:
| Column | What it records |
|---|---|
| Section | Which part of the paper (Introduction, Methods, etc.) |
| Unit purpose | What this section does for the argument |
| Motivation link | How it serves the confirmed central motivation |
| SOTA lessons | What was learned from the example papers that informed this section’s approach |
| Evidence support | Which items from the evidence bank underpin the claims here |
| Text checks | Specific checks the draft must pass (no unsupported claims, venue tone match, etc.) |
This matrix is the audit trail. A coauthor, advisor, or reviewer can read it and immediately see whether the paper is arguing coherently or drifting. It is also the primary input for revision: when a reviewer asks you to reframe the contribution, you update the matrix first, then the text.
Research Phase: Flash vs. Pro
Before any writing, PaperSpine runs a structured literature survey:
| Mode | Papers surveyed | Use when |
|---|---|---|
| flash | 3 SOTA examples + 3 recent papers | Time-limited; quick positioning |
| pro | 6 SOTA examples + 6 recent papers | Full submission; want deep venue calibration |
The output feeds directly into the evidence bank and informs which claims the rationale matrix will flag as requiring citation.
Evidence Bank and Claim Register
No claim enters the draft without a corresponding entry in the claim register. The claim register tracks:
- The claim itself
- The evidence bank item(s) that support it
- The section where it appears
- Whether it has been verified against a fetched or provided source
This is the structural answer to hallucination: not a prompt that says “only make claims you can support,” but an artifact that makes unsupported claims visible before submission.
Output Formats
| Format | Notes |
|---|---|
main.tex + .bib |
Always produced; journal/conference-ready LaTeX |
.docx |
Optional; for advisors or coauthors who don’t use LaTeX |
| Optional; compiled from LaTeX | |
| Chinese translation package | Optional; full CN translation of the finished paper |
Installation and Invocation
Codex:
$paper-spine
Bundled skill in dist/codex/paper-spine/. Available after plugin install.
Claude Code:
/paperspine
Flat skill suite in dist/claude/. Install via:
/plugin marketplace add https://github.com/WUBING2023/PaperSpine
/plugin install paper-spine
/reload-plugins
Both environments run the same underlying workflow; the invocation surface adapts to the tool.
How It Differs from academic-research-skills (Imbad0202)
These are complementary academic writing suites, not overlapping:
| Dimension | academic-research-skills (Imbad0202) | PaperSpine (WUBING2023) |
|---|---|---|
| Primary focus | Research workflow — finding, organizing, synthesizing literature | Manuscript construction — writing, rationale, revision |
| Entry point | A research question | A motivation or existing draft |
| Signature mechanism | Socratic intent detection + PRISMA systematic review | Writing rationale matrix + claim register |
| Review simulation | Yes (EIC + 3 reviewers + Devil’s Advocate) | No |
| LaTeX output | Yes (via academic-paper skill) | Yes (core output) |
| Evidence tracking | Via systematic review pipeline | Via evidence bank + claim register |
| Best for | Researchers who need to master the literature before writing | Writers who have the material and need to construct the argument |
The ideal pipeline for a serious paper: use academic-research-skills to master the literature, then hand the evidence to PaperSpine to construct the manuscript.
How LearnAI Team Could Use This
- Paper writing module for grad students — PaperSpine’s motivation-first discipline maps directly onto the advice faculty give in every thesis seminar. Teaching it as a skill puts a structure around advice that is usually informal (“make sure your paper has a clear argument”).
- Undergraduate writing quality — Use the rationale matrix as a grading scaffold: does each section justify its existence against the paper’s central claim? Students who use PaperSpine will naturally produce papers that answer this question.
- Research methods courses — The evidence bank + claim register is a practical implementation of the “every claim needs a citation” rule. Assign it as infrastructure, not just as a style guide.
- Academic integrity framing — PaperSpine is a strong example of AI-augmented writing that keeps the researcher in the epistemic loop. The model drafts; the human controls the motivation, verifies the evidence, and owns the argument. Use it to demonstrate responsible AI use in research.
- Conference/journal submission preparation — The flash vs. pro research modes map onto time constraints real researchers face. A short workshop paper can run flash; a journal submission should run pro.
Real-World Use Cases
| Scenario | How to use |
|---|---|
| Rejected manuscript revision | Use “Rewrite Existing” workflow; the rationale matrix will surface structural problems a reviewer reacted to but couldn’t articulate |
| PhD thesis chapter | Confirm motivation at the chapter level; use the matrix to ensure the chapter argument is coherent before drafting 20,000 words |
| Conference deadline crunch | Run “flash” research mode; limit the claim register to the 5–7 claims the paper absolutely must make; draft to the matrix |
| Collaborative paper with international coauthors | The rationale matrix serves as a shared document all coauthors can read and challenge before writing begins, reducing revision cycles |
| Course research paper (undergraduate) | Lower the bar: use it with 2 SOTA + 2 recent papers; the claim register teaches citation discipline as a workflow, not just a rule |
| Multi-language submission | Use the Chinese translation package for papers targeting Chinese venues without a separate translation step |
Important Things to Know
- Very new. Created May 17, 2026 — ~240 stars (as of May 2026) in days, which means the workflow is less battle-tested than academic-research-skills (which has 18K+ stars). Expect rough edges.
- Motivation-first is non-negotiable. The skill won’t bypass the
confirmed_motivation.mdstep. This is by design. If a student or researcher finds this frustrating, that frustration is itself a signal worth examining. - The claim register does not fetch sources for you. It tracks claims and their evidence pointers; you still need to supply the actual papers or use academic-research-skills to surface them.
- LaTeX is the primary output. The DOCX option exists but is secondary. Teams that live in Word or Google Docs should pair this with a LaTeX-to-DOCX conversion step.
- No peer review simulation. Unlike Imbad0202’s suite, PaperSpine does not simulate reviewer feedback. Run the finished manuscript through academic-research-skills’
academic-paper-reviewerif you want a pre-submission review pass. - License: MIT. The repo is MIT-licensed, so reuse and adaptation are permitted with attribution.
Links
- GitHub: WUBING2023/PaperSpine
- Companion entries: Academic Research Skills for Claude Code, AI Agent Academic Research Writing, Posterskill