Academic Paper Orchestrator

Use this skill to coordinate a complete academic paper workflow. Do not do every specialized task inside this skill; route to the focused skill that owns the current stage.

Routing Map

If a focused skill is unavailable, follow the same ownership boundaries and tell the user which part was handled without the dedicated skill.

Core Rules

1. Ask only the minimum startup questions needed to choose the next stage.
2. Maintain visible project state, current gate, confirmed artifacts, missing materials, risks, and next step.
3. Treat verified evidence as the source of truth for scholarly claims.
4. Treat real codebase, dataset, or experiment artifacts as the source of truth for implementation or result claims.
5. Keep research, parsing, writing, figures, formatting, and post-draft review as separate passes.
6. Stop for user confirmation at required gates before moving to the next major stage.
7. Never fabricate citations, DOI records, experiment results, repository behavior, figure contents, or formatting rules.
8. Stop main-text or rebuttal writing when required support is missing; request the missing material before continuing.
9. Treat formulas, superscripts/subscripts, citations, cross-references, figures, and tables as conversion-sensitive artifacts.
10. For substantial paper workflows, plan multi-agent parallel review, verification, QA, or action-plan decomposition by default, but start sub-agents only after user confirmation unless prior full or automatic-execution permission exists.
11. For combined post-draft quality and language review, run quality review first, language-style review second, then send the unified priority queue to $revision-control.

Intake

Collect only what is missing:

• paper type, language target, and delivery target;
• research field and target venue/school field when terminology may matter;
• target standard: school handbook, Word template, journal guide, supervisor requirement, or provisional baseline;
• target examples: 3-10 accepted papers, approved theses, templates, author guidelines, or supervisor-approved chapters when available;
• current materials: topic, outline, draft, PDFs, DOCX files, data, codebase, figures, bibliography, and notes;
• constraints: deadline, citation style, word/page count, required chapters, and permitted tools.

Before routing work into execution, ask for materials that would materially improve quality:

Required or high-value materials:
Missing materials:
Can proceed without them: yes/no
Fallback if unavailable:

Workflow

1. Rebuild or create the project state.
2. Decide the current mode and route using the routing map.
3. Apply workflow-protocol-index.md and trigger-conflict-matrix.md when a request spans stages or the owner skill is ambiguous.
4. Apply multi-agent-academic-workflow-gate.md before substantial review, verification, QA, or reviewer-comment decomposition work that can be split into independent checks.
5. Apply orchestration-contract.md for substantial projects, state files, material passports, and handoff packets.
6. Apply project-gates-and-phases.md for full-project phase sequencing and required gates.
7. Apply integrity-gate-patterns.md before drafting, benchmark review, final polish, formatting, or any claim about code/data/results.
8. Apply writing-chain-gates.md when the project is entering field confirmation, evidence-gap handling, reviewer-comment response, post-draft review, or final polish.
9. Apply main-text-and-rebuttal-claim-support-gate.md before final main-text or rebuttal claims.
10. Apply reviewer-comment-action-plan-gate.md when two or more reviewer, editor, committee, or supervisor comments are provided.
11. Apply notation-and-conversion-integrity-gate.md before lossy manuscript, figure, table, formula, citation, or cross-reference conversion.
12. Apply dependency-and-software-guide.md when a stage may need Pandoc, Word, PDF/OCR tools, draw.io, Matplotlib, Graphviz, Inkscape, network APIs, or project-specific Python packages.
13. Route execution to the focused skill that owns the current stage.
14. When quality review and language review are both requested, sequence $post-manuscript-benchmark-review then $language-style-review, mark suggestions that could roll back quality priorities, and hand the confirmed queue to $revision-control.
15. At each gate, report completed work, artifacts, decisions, risks, missing materials, and recommended next step.

For narrow requests, run only the relevant stage and state what was intentionally skipped.

Evidence And Version Discipline

• Keep a single evidence register or literature master list for the project.
• Mark items as candidate, verified, downloaded, parsed, cited, rejected, or unresolved.
• Track major project inputs with material passports when multiple stages depend on them.
• Keep paired Markdown and DOCX outputs at the same semantic version when both exist.
• Save new outputs with explicit suffixes; do not overwrite the only working draft unless the user explicitly requests it.
• Validate project_state.json before moving from research/parsing into drafting, and before moving from writing into formatting.

Fallbacks

• If literature verification requires network access and network is unavailable, create a manual verification queue with exact DOI/title/search strings.
• If PDF parsing fails, list failed pages and recommended manual recovery.
• If a DOCX needs Word-only features, explain what can be automated and what requires Word.
• If a requested figure cannot be drawn faithfully from evidence or code, decline the specific figure or mark it unresolved.
• If the user wants to skip a gate, comply only after stating the risk.

Reference

Read stage-map.md when a full project needs a compact stage checklist or when resuming a long-running thesis workflow.

Read workflow-protocol-index.md before selecting shared protocol references for cross-skill coordination.

Read multi-agent-academic-workflow-gate.md when planning default parallel checks for substantial paper workflows, including literature verification, draft review, sentence-level language review, figure/format QA, or reviewer-comment action decomposition.

Read trigger-conflict-matrix.md when a request could trigger multiple academic skills or when deciding whether the orchestrator should own the task.

Read revision-control-contract.md when routing formal revision work, sentence pass/fail state, object libraries, project standards, or upgrade plans.

Read language-style-review-schema.md when consuming a language-style review report or merging it with quality review findings.

Read handoff-field-schema.md when checking field names for workflow modes, material passports, claim anchors, literature gaps, writing-ready handoffs, or benchmark reports.

Read main-text-and-rebuttal-claim-support-gate.md before drafting, revising, polishing, or approving supported main-text or rebuttal claims.

Read reviewer-comment-action-plan-gate.md when a comment batch needs triage and concrete revision actions before editing.

Read notation-and-conversion-integrity-gate.md when formulas, superscripts/subscripts, citations, cross-references, figures, tables, or captions are converted across formats.

Read project-artifact-templates.md when creating a project state file, handoff packet, or target-baseline note.

Read orchestration-contract.md for project-state output, route plan, handoff packets, and state-file use.

Read project-gates-and-phases.md for standard phases, required gates, and stage execution notes.

Read integrity-gate-patterns.md before accepting claims about sources, code, data, experiments, benchmarks, figures, or final deliverables.

Read writing-chain-gates.md for field confirmation, LIT_GAP handling, reviewer-comment revision planning, post-draft review routing, and final de-AI polishing gates.

Read dependency-and-software-guide.md when a workflow stage may require external software, Python packages, Word automation, Pandoc, draw.io, image conversion tools, or network APIs.

Bundled Utilities

• scripts/project_state_check.py: create a JSON project-state template or validate project state, required handoff packets, gate consistency, evidence readiness, and formatting readiness.