Research To Paper

Research-to-Paper (orchestrator)

Entry point for a self-contained workflow. Nothing here calls another plugin — search, verification, review, writing, de-AI, and format rendering all live in this skill's six sub-skills, so it works standalone (a collaborator installs this one plugin and has everything). Find which stage the user is at, run the stages they need in order, and pass each stage's output into the next.

The stages

1. SCOPE — decide the direction → research-to-paper-scope.

Reads authoritative literature, then confirms an angle, scope, target journal, and word count one step at a time, and looks up the journal's author requirements once chosen. Output: scope_brief.md. Skip only if the user already has a confirmed direction.

1. CURATE — build a verified library → research-to-paper-curate.

Built-in five-source search, per-paper CrossRef DOI verification, a multi-agent adversarial review gate, then export to RIS (Zotero/EndNote) + BibTeX + a by-category Excel. The SCOPE themes become the categories.

1. WRITE — draft, review, de-AI → research-to-paper-write, which orchestrates two companion sub-skills:

research-to-paper-audit (independent multi-reviewer rounds) and research-to-paper-humanize (five-dimension de-AI incl. 长短句). Understands the content first, plans a per-unit rationale, drafts with evidence-matched hedging, then audits until clean and de-AIs — discussing with the user each round. Scenes: journal (论文) / conference (会议论文) / report (报告) / review (综述) / competition (竞赛).

1. BUILD — render to formats → research-to-paper-build.

Turns the approved manuscript into LaTeX (.tex), Word (.docx), and optionally PDF via pandoc, resolving citations from the library.bib.

How to route

• Vague idea → SCOPE → CURATE → WRITE → BUILD.
• Has a direction, needs references → CURATE → WRITE → BUILD.
• Has direction + library, wants to write → WRITE → BUILD.
• "Verify these DOIs" / "export to EndNote" → CURATE.
• "Help me figure out the angle / which journal" → SCOPE.
• "Draft / rewrite this" → WRITE.
• "Review / audit this draft" / "查无据claim" → research-to-paper-audit.
• "降AI率 / de-AI this" / "长短句" → research-to-paper-humanize.
• "导出 LaTeX/Word/PDF" / "转成 docx" → research-to-paper-build.
• "产物齐了吗 / 检查产物 / verify the artifacts" → scripts/artifact_check.py.
• "引用对得上吗 / 引用能解析吗 / check the citations resolve" → scripts/cite_check.py.

Each stage is independently useful, so don't force the full pipeline when the user asked for one piece. Hand off cleanly and tell the user what each produced before moving on.

Workspace & artifacts

Run a multi-stage job inside one working directory so the reasoning trail is auditable, not just the final file. Scaffold it once at the start:

python scripts/new_workspace.py manuscript_workspace

Each stage writes its artifact into this workspace — scope_brief.md, library/, writing_rationale_matrix.md, draft.md, audit_report.md, humanize_matrix.md, final/. The full list (what each records) and the verification commands are in references/artifacts.md. Before calling a run done, check completeness and that every citation resolves:

python scripts/artifact_check.py manuscript_workspace --write   # refreshes artifact_manifest.md
python scripts/cite_check.py manuscript_workspace/draft.md manuscript_workspace/library/library.bib

A polished draft with an empty rationale matrix or no audit report is an incomplete run, not a finished one.

Files

• scripts/new_workspace.py — scaffold the manuscript_workspace/ and seed scope_brief.md.
• scripts/artifact_check.py — verify the artifact trail; write artifact_manifest.md.
• scripts/cite_check.py — check draft [@key] / \cite{} citations resolve against library.bib.
• references/artifacts.md — the workspace layout, the full artifact set, and every check command.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Jason-0409-G
• Source: Jason-0409-G/scriptorium
• License: MIT

Install and usage instructions live in the source repository linked above.