mcpField
P
paper-spine-rewrite

paper-spine-rewrite

github.com/wubing2023Documents6.1k353 installsUpdated 2w ago
Repository

About paper-spine-rewrite

A PaperSpine skill for substantively revising an existing academic manuscript using research, confirmed motivation, evidence, and a documented writing plan. It requires prerequisite research and citation materials before rewriting.

Key features

  1. 01Maps original manuscript logic
  2. 02Builds evidence and section plans
  3. 03Documents rewrite decisions
  4. 04Checks integrity and review requirements

Use cases

  1. 01Restructuring a draft around confirmed motivation
  2. 02Revising claims against available evidence
  3. 03Preparing manuscript sections for final review

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/wubing2023/paperspine --skill paper-spine-rewrite

After installing, type/paper-spine-rewritein Claude Code to confirm it loaded.

Bundled files

References and scripts installed along with SKILL.md.

agents/
openai.yaml
references/
deep-imitation-protocol.md
logic-transfer-audit.md
motivation-thread-writing.md
rewrite-matrix.md
round1-literature-revision.md
round2-journal-revision.md
version-requirements.md
writing-rationale-matrix.md

Details

From the project README

PaperSpine Rewrite

Use this skill when the user already has a draft and wants a substantive manuscript improvement.

Required Inputs

  • paper_rewriting_output/paper_spine_config.json
  • user draft from draft_path or the conversation
  • paper_rewriting_output/reference_materials/source_index.md
  • paper_rewriting_output/research_dossier.md
  • paper_rewriting_output/exemplar_learning_dossier.md
  • paper_rewriting_output/style_profile.md
  • paper_rewriting_output/sota_gap_map.md
  • paper_rewriting_output/citation_support_bank.md
  • paper_rewriting_output/confirmed_motivation.md with user confirmation
  • evidence from the user's draft, figures, tables, data, or notes

If research or confirmed motivation is missing, do not rewrite. Return to paper-spine-research and ask the user to confirm the motivation after seeing research-grounded options.

If citation_support_bank.md is missing or shallow, return to paper-spine-citation before final writing. Introduction, background, related work, and Discussion claims should draw from that bank when they need literature support.

Required Outputs

  • paper_rewriting_output/original_logic_map.md
  • paper_rewriting_output/evidence_bank.md
  • paper_rewriting_output/section_blueprints.md
  • paper_rewriting_output/writing_rationale_matrix.md
  • paper_rewriting_output/rewrite_matrix.md
  • paper_rewriting_output/logic_transfer_audit.md
  • revised manuscript or revised sections
  • translation_zh/ package when translation_package is zh (via paper-spine-translate)

Original Logic Map

Before rewriting, map the existing manuscript in order:

Original UnitCurrent Text RoleEvidence UsedMotivation LinkProblemKeep / Move / Rewrite / Delete

This is required so the rewrite can be compared against the original logic, not only against surface wording.

Writing Rationale Matrix

Before final prose, create writing_rationale_matrix.md and read references/writing-rationale-matrix.md. The matrix is the rewrite plan. It must be ordered by the manuscript's actual writing units, not by a fixed IMRaD or journal-paper template.

Use this table:

Row IDManuscript UnitOriginal Problem or Planned FunctionMotivation LinkReference/SOTA Pattern LearnedTarget Scene or Venue NormUser Evidence or Citation AnchorPlanned ChangeFinal Text Check

The first data row must deeply justify the whole-work framework or throughline: why this controlling structure is chosen, how SOTA/target examples informed it, how it follows the confirmed motivation, which user evidence anchors it, and how the final manuscript will be checked against it. After that, split the draft into the smallest useful units for this target scene: paragraphs, paragraph groups, result/claim units, model steps, assumptions, review synthesis blocks, competition solution blocks, headings, captions, or other argument-bearing fragments. Do not force reports, reviews, or competition papers into Abstract/Introduction/Methods/Results/Discussion if that is not the right structure.

Each row must explain concrete anchors across multiple dimensions. Acceptable reasons include: confirms or narrows the central motivation, transfers a structural move from a strong example without copying wording, matches target-scene expectation, moves evidence next to the claim, fixes a weak transition, creates a front/back echo, or constrains a claim to available evidence. If a row cannot teach the user why this writing move is better, it is too shallow.

Humanize Tier

If paper_spine_config.json has humanize_tier set to light, medium, or heavy, route to paper-spine-humanize before writing. The humanize skill provides tier-specific writing constraints enforced during all prose generation.

Rewrite Rules

  • Rewrite from writing_rationale_matrix.md, not by appending sentences to old paragraphs.
  • A paragraph should survive unchanged only if the matrix explicitly says why it already serves the confirmed motivation.
  • Preserve user claims only when supported by user evidence.
  • Preserve LaTeX commands, labels, citations, equations, figures, and tables unless the user asks to change structure.
  • Use output_language from config: en for English, zh for Chinese.
  • Select citations from citation_support_bank.md only when they support a specific sentence. Do not add citation clusters without a clear claim.
  • Keep target-scene style subordinate to the confirmed motivation.
  • rewrite_matrix.md must map original units to final units and state whether each change is structural, rhetorical, evidence-related, or only language.

Pre-LaTeX Gate

Before routing output to paper-spine-latex, run:

python scripts/integrity_audit.py paper_rewriting_output --markdown --write
python scripts/structured_review.py paper_rewriting_output --dispatch

integrity_audit.md must show no BLOCKED findings. After dispatch, launch three parallel review sub-agents per review_prompts/dispatch.md. Validate independence with structured_review.py --validate review_prompts.

Read references/writing-rationale-matrix.md, references/motivation-thread-writing.md, references/deep-imitation-protocol.md, and references/rewrite-matrix.md when doing substantive rewriting.

Comments

Markdown supported