Paper Spine

PaperSpine For Codex

Use this skill for any PaperSpine task in Codex. This is the Codex edition: one official-style skill folder with all PaperSpine workflow logic, references, and scripts bundled inside. Do not depend on separate child skills in Codex.

Core Rule

PaperSpine is not a prose polisher. It must research the target scene, learn strong examples, confirm the controlling motivation with the user, design the writing unit by unit, and only then write, rewrite, build LaTeX, translate, and audit.

Never fabricate data, metrics, citations, figures, experiments, or claims. External papers teach structure and rhetoric only; user materials are the only source for this manuscript's evidence.

Mandatory Intake

Before starting a workflow, create config through the terminal wizard whenever possible. In Claude Code, use /paperspine; it automatically launches the intake UI when configuration is missing. scripts/launch_paperspine_ui.ps1 is the underlying launcher for the real interactive terminal window. The terminal UI supports Up/Down option switching, Left/Right field switching, Enter edit/confirm, and S save. Do not run the Python input() wizard inside hidden Bash/tool execution.

python scripts/intake_wizard.py --output-dir paper_rewriting_output

The wizard must collect all config fields in the command line by numbered menus or text input, show a final review screen, and allow editing before writing the config. Do not ask the user to manually write JSON or answer a long plain chat checklist unless the terminal is unavailable.

On first setup or when the user asks to change global preferences, run:

python scripts/intake_wizard.py --setup-global --output-dir paper_rewriting_output

Global setup records the command-line UI language preference (zh or en).

Required config fields:

Workflow

1. Run intake wizard and write paper_spine_config.json and .md.
2. Create or verify source_map.md.
3. Research before final motivation. Create reference_materials/, research_dossier.md, exemplar_learning_dossier.md, style_profile.md, sota_gap_map.md, and motivation_options_after_research.md.
4. Stop for user confirmation. Write confirmed_motivation.md only after the user chooses or revises the motivation.
5. For rewrite_existing, create original_logic_map.md, evidence_bank.md, section_blueprints.md, writing_rationale_matrix.md, rewrite_matrix.md, and logic_transfer_audit.md before final prose.
6. For build_from_materials, create source_inventory.md, evidence_bank.md, figure_asset_map.md, claim_register.md, section_blueprints.md, and writing_rationale_matrix.md before final prose.
7. Always produce final_paper/main.tex. Compile paper.pdf when a TeX engine is available. Markdown alone is not final.
8. If word_output=docx, produce final_paper/paper.docx and word_report.md.
9. If output_language=en and translation_package=zh, produce the full translation_zh/ package. This means complete file-by-file translation of every required intermediate Markdown artifact, including large tables such as writing_rationale_matrix.md, plus a complete Chinese full-paper translation.
10. Run artifact_check.py --markdown --write before completion.

Writing Rationale Matrix

writing_rationale_matrix.md is the writing plan, not a change log. It must be much more detailed than simple rows such as "fix graphicspath" or "rewrite abstract".

Use this table:

Rows must follow the real manuscript/report order and split the work into paragraph-sized, claim-sized, evidence, model, synthesis, heading, caption, or competition-solution units. A row should teach the user why the writing move is better: what logic failed, what strong examples do, why the target venue expects it, what evidence supports it, and how the final text should prove the claim.

The first row must be a deep whole-work framework row. It should explain the chosen controlling structure, how SOTA/target examples informed that structure, how the confirmed motivation orders later sections, which user evidence anchors the arc, and how the final manuscript will be checked against it. Do not write a one-line "conservation-first" or "problem-solution arc" summary without this reasoning.

A generic row fails. Examples of failing reasons: "improve clarity", "polish", "fix path", or "make abstract conservation-first" without detailed logic.

Translation Package

When translation_package=zh, translate all required intermediate Markdown artifacts and also produce:

• translation_zh/writing_rationale_matrix.zh.md: complete row-by-row Chinese translation of writing_rationale_matrix.md; preserve the full table and row count.
• translation_zh/full_paper_translation.zh.md: complete Chinese translation of the final paper text, including title, abstract, section prose, captions, table notes, limitations, and conclusion. Preserve citation keys, labels, formulas, file paths, and raw numeric values.
• translation_zh/translation_coverage.md: row-by-row coverage of every source artifact and every final-paper section/caption/table note.

Partial translation is a failed audit.

References To Read As Needed

• references/interactive-intake.md for config and wizard behavior.
• references/flash-pro-research.md and references/scenario-*.md for target research.
• references/writing-rationale-matrix.md for detailed unit-by-unit planning.
• references/build-from-materials.md for from-zero construction.
• references/rewrite-matrix.md and references/logic-transfer-audit.md for rewrite work.
• references/translation-package.md for complete Chinese translation output.

Scripts

Prefer bundled scripts over ad-hoc checks:

python scripts/intake_wizard.py --output-dir paper_rewriting_output
python scripts/material_inventory.py <materials_dir> --output-dir paper_rewriting_output
python scripts/artifact_check.py paper_rewriting_output --markdown --write
python scripts/latex_guard.py final_paper/main.tex --markdown
python scripts/word_guard.py final_paper/paper.docx --markdown
python scripts/revision_audit.py original.tex revised.tex --markdown