Paper Workflow Orchestrator

Paper Workflow Orchestrator

Use this skill to plan and coordinate manuscript work. Do not use it as a prose-polishing skill; hand off paragraph-level editing to scientific-writing, scientific-prose, or writing-fidelity.

Research Paper Production Workbench

Use this skill when the user needs the paper process itself organized, not just a paragraph rewritten. Route by task shape:

1. Intake and scoping: identify the research question, target audience, current artifact state, available evidence, missing evidence, and deadline pressure.
2. Argument spine: turn scattered notes, experiments, or conclusions into a problem -> gap -> method -> evidence -> claim -> limitation structure.
3. Section contracts: define what each manuscript section must prove, which figures/tables carry the proof, and which claims must be removed or narrowed.
4. Draft production: sequence the writing so methods, results, figures, and claims stay synchronized; do not generate a polished paper that outruns the evidence ledger.
5. Revision loop: audit reviewer attack surfaces, unsupported superiority or robustness claims, missing comparisons, stale numbers, and inconsistent terminology before prose polishing.
6. Submission/rebuttal handoff: create venue checklist, reviewer concern matrix, required experiments or text fixes, and final artifact QA tasks for the downstream writing, citation, LaTeX, PDF, or review skills.
7. Result-to-claim gate: after experiments or audits finish, decide which claims are supported, which must be narrowed, which are draft-only, and which require new evidence before writing.
8. Final artifact gate: before treating a manuscript, Markdown, PDF, response, or report as final, identify the selected artifact, evidence authority, unresolved checks, and the downstream style/fidelity skill that must run.

Use literature-review for single-paper cards or field synthesis. Use peer-review for reviewer scoring, acceptance risk, and rebuttal assessment. Use writing-fidelity before final Markdown/PDF delivery when version labels, headings, numbers, or evidence boundaries have been corrected before.

Workflow

1. Establish the paper type, venue, audience, artifact state, target outputs, and hard constraints.
2. Create a claim/evidence spine:
   • central contribution;
   • supporting claims;
   • evidence artifacts for each claim;
   • figure/table panels that carry each claim;
   • missing or weak evidence.
3. Choose the next active phase:
   • bootstrap directory and source files;
   • outline and section plan;
   • draft section;
   • revise claim/evidence alignment;
   • audit citations and figures;
   • prepare submission package;
   • respond to reviewers.
4. Keep one source of truth for manuscript state. If the repo already has a paper plan, update or reference it instead of creating a parallel plan.
5. Run a result-to-claim check before drafting or accepting results prose: each number, table row, leaderboard entry, audit result, and qualitative example must be classified as supported, bounded, draft-only, historical, or not verifiable.
6. Before final reporting, state what changed, what evidence supports it, what remains unresolved, and which downstream skill should run next.

Routing

• Use scientific-writing for full paragraph scientific prose.
• Use literature-review for field synthesis and related-work structure.
• Use citation-verification before submission or when claims rely on citations.
• Use nature-manuscript-workflow for broad-journal/high-impact framing, figure-to-claim logic, data availability, and Nature-family checks when relevant.
• Use latex-paper-authoring when LaTeX structure, Overleaf readiness, or compilation is the main issue.
• Use scientific-visualization for manuscript figure palettes, plotting snippets, figure export QA, top-conference figure presets, and publication-ready visual checks. Use drawio-diagrams or d2-diagrams when editable diagrams are the main deliverable.

Quality Gates

• Every major section must have a purpose in the argument.
• Every figure panel must map to a specific claim or method step.
• Every central claim must identify the artifact that supports it.
• Missing evidence must be named explicitly, not hidden by prose.
• Do not invent results, citations, reviewer requests, or venue policies.
• Do not let an audit report, old draft, month label, or candidate row substitute for the author-approved best result or the requested final artifact.