research-workflow-orchestrator

name	research-workflow-orchestrator
description	Use when coordinating a full research project, graduation thesis, dissertation, paper pipeline, literature-to-experiment-to-writing workflow, or when deciding which research, notebook, PDF, DOCX, citation, visualization, or audit skill to use next.

Research Workflow Orchestrator

Use this skill as the entry point for a multi-stage research or thesis workflow. It coordinates specialized skills instead of replacing them.

Core Rule

Always turn a broad research request into a staged workflow with artifacts, skill sequence, risks, and next concrete actions. Prefer real project files and evidence over generic advice.

Skill Routing

• Planning and advisor topic intake: use $research-paper-plan.
• Title-first thesis control: when the project starts from a fixed or tentative title, use project-scope-control.md to track title survival, causal availability, node/structure definition, and downgrade/rename policy across stages 1, 3, 4, 8, and 10.
• Idea discovery and early research direction: use $research-paper-plan with docs/thesis/idea-discovery.md for paper pools, idea matrices, novelty risks, and shortlists.
• Literature discovery, Zotero literature hub management, source-grounded paper reading, related work, section-level citation matching, deep research task packets, Zotero screening loops, and citation batches: use $research-literature-review; use $semanticscholar-skill for real Semantic Scholar searches when exact papers, citations, or BibTeX candidates matter; use $pdf when full-paper PDF reading, original/translation alignment, figure/table extraction, or source anchors matter.
• Experiment architecture, code design, local_mac CPU-only smoke tests, remote_desktop_4060 primary GPU runs, cloud_autodl fallback training with evidence auto-save and auto-shutdown, training/evaluation scripts, config/log/output conventions, reproducibility, and mapping code runs to thesis experiments: use $research-experiment-engineering; use $research-code-quality for skeletons, contract checks, smoke configs, output manifests, 4060 handoff templates, and AutoDL fallback templates.
• Human-supervised iterative experiment improvement: use $research-autoresearch-loop for autoresearch-results.tsv, autoresearch-state.json, verify gates, guard gates, baseline comparison reports, and recovery decisions.
• Experiment recording and exploration: use $jupyter-notebook.
• Result interpretation: use $research-results-analysis; use Build Web Data Visualization as a chart-choice, uncertainty-display, dashboard-chart, evidence-graph, accessibility, and visual-QA guide when results need to become advisor-facing visuals.
• Figures, tables, diagrams, captions, Nature-style figure audit, and model-architecture rendering: use $research-paper-figures; for thesis route diagrams, method architecture diagrams, mechanism diagrams, workflow diagrams, and system-structure diagrams, prefer academic PPT templates followed by PPTX-safe editing, thumbnail review, WPS/PowerPoint manual polish, and final export. Image Gen is optional inspiration only, not the default mother draft. Use draw.io as a fallback when PPT templates are unsuitable or an editable vector source is required. On Windows with Microsoft Visio, optionally use the Visio JSON-plan route inspired by codex-visio-replica-workflow. Use Build Web Data Visualization for complex chart selection and dashboard/evidence graph design QA.
• Drafting and revision: use $research-paper-writing.
• Data provenance and data availability: use $research-data-availability before final writing, final audit, DOCX/PDF production, and defense preparation.
• DOCX thesis/manuscript production: use the Documents plugin (documents:documents in Codex plugin contexts). Pages or Microsoft Word may be used locally for human review, but Word is optional.
• PDF review or final rendering checks: use $pdf.
• Final submission or defense checks: use $research-final-audit.
• Task planning and progress dashboards: use Notion when the user wants project tasks, weekly plans, or supervisor feedback synced outside the repo.
• Code versioning and GitHub work: use GitHub for repository state, issues, PRs, CI, and version traceability around major code changes or formal experiments.
• CodeRabbit: optional pre-merge AI code review for important scripts, dashboard, CI, and skill changes. Run only when authenticated, and do not treat it as a default CI dependency.
• Plugin gates: use plugin-gate-policy.md, plugin-review-log.md, and scripts/plugin_gate_advisor.py to recommend Codex Security, Build Web Apps, Data Analytics, Product Design, or CodeRabbit checks by stage. These gates are review aids, not evidence sources.
• Codex Security: use when stages 5-6 touch local Dashboard APIs, file-writing scripts, remote 4060 scripts, CI, open-path boundaries, or other security-sensitive code.
• Build Web Apps: use for Dashboard React/Vite implementation, interaction bugs, responsive layout, Browser/Safari checks, and frontend QA.
• Data Analytics: use in stages 7-8 for result data quality, metric diagnostics, baseline-delta, uncertainty, failed-run, and anomaly checks.
• Product Design: use in stages 9 and 12 for advisor-facing Dashboard, figure, diagram, and defense-slide readability or visual hierarchy review.
• Device workflow: treat local_mac as the stages 1-11 research console and main DOCX/PDF/PPTX production machine, remote_desktop_4060 as the primary GPU execution target, cloud_autodl as a stronger fallback, and the Windows laptop as the stage 12 compatibility review and final submission-check machine.
• Evidence workflow: use evidence-promotion-policy.md and id-lifecycle-policy.md to keep SEC-*, SEG-*, CLM-*, EXP-*, DATA-*, FIG-*, MAT-*, CIT-*, BMK-*, ZCOL-*, DRT-*, and ZREV-* relationships consistent before final writing or defense.
• Project overview: use workflow-dashboard.md, project-scope-control.md, console-file-index.md, weekly-review.md, the legacy-compatible daily-workflow-entry.md, scripts/research_workflow_doctor.py, scripts/audit_project_scope.py, scripts/update_daily_workflow.py, scripts/update_weekly_review.py, scripts/export_evidence_graph.py, scripts/dashboard_control_server.py, scripts/edit_workflow_record.py, and dashboard-web/ when the user needs a one-page status view, title-scope control, file-layer guidance, weekly convergence, current research workspace, health check, focused evidence chain, experiment comparison, or local web dashboard with refresh/open/export/edit actions.
• Read-only dashboard publishing: Vercel may be considered later for an advisor/demo preview, but the local write API, private research data, and unpublished thesis evidence must not be deployed.
• Skill maintenance: use scripts/audit_skills.py when changing or installing skills, or before merging workflow changes, to catch missing references, missing scripts, and old tool assumptions.
• Engineering discipline: use Superpowers for TDD, systematic debugging, verification, and code review during implementation work.
• Spreadsheet exports: use Spreadsheets for reviewable literature, result, claim, and audit tables without replacing source records.
• Defense slides: use Presentations for PPTX creation/editing; use PPT template figures or draw.io exports for structured diagrams; use Figma or BioRender only as optional visual refinement tools when available. Canva is optional only if explicitly available and useful.
• PPT template editing / PPTX-safe editing is the default stage 9 route for method overview, technical route, workflow, mechanism, and system-structure figures when a suitable academic template exists. The workflow is: choose template -> fill thesis-specific Chinese labels -> inspect thumbnails -> apply safe object/XML edits only when needed -> WPS/PowerPoint manual polish -> export SVG/PDF/PNG/PPTX.
• Image Gen Skill is no longer mandatory for stage 9 method/architecture/route figures. It may be used only as optional inspiration when no suitable template/reference exists; its output is checked for content accuracy and must not be treated as final manuscript evidence.
• draw.io / draw.io MCP is a fallback stage 9 formal redraw route for model architecture, evidence graph, and system diagrams when PPT templates are unsuitable or editable vector structure is required. Export SVG/PDF/PNG for thesis use and package into PPTX for defense slides when needed.
• Windows Visio is an optional direct-use route for figure replication: reference image -> JSON plan -> .vsdx -> EMF/PDF/PNG -> thesis/PPTX handoff. Record it in diagram-replica-tasks.md; do not let Visio become the source of truth.
• Zotero literature hub is the stage 2-10 citation backbone: Zotero stores the local library, zotero-literature-hub.md stores the inventory/policy snapshot, zotero-screening-loop.md records intake decisions, zotero-collection-coverage.md checks collection coverage, and citation-provenance.md promotes formal CIT-* rows. Do not cite unverified candidates directly from Zotero.
• Figma can be used in stage 9 or 12 for optional diagram refinement, reusable design components, and visual system work after the draw.io/source-of-truth record and visual reference are established.
• Scite and Zotero are the default citation-verification helpers: Scite checks support/contrast/mention status, while Zotero manages local library records and citation exports. Neither replaces direct source-grounded reading for critical claims.
• Supabase database migration, Overleaf synchronization, reviewer-response workflow, and arXiv auto-submit are out of scope for the current graduation-thesis workflow.
• Nature-derived high-standard enhancements: use downloaded nature-academic-search, nature-figure, nature-reader, nature-citation, nature-data, nature-writing, nature-polishing, and nature-paper2ppt source material as reference layers. nature-academic-search strengthens multi-source search routing, nature-figure strengthens figure audit, nature-reader strengthens source-grounded paper reading, nature-citation strengthens long-text citation batching, nature-data strengthens data availability, nature-writing/polishing strengthens final prose and advisor precheck, and nature-paper2ppt strengthens defense slides. They do not replace the local research-* evidence workflow. Do not add nature-response for graduation-thesis-only workflows unless the user later asks for journal review response support.

Project Console

Project Template Initialization

Use the global template source when a project needs to be initialized or repaired:

<research-workflow-kit>/
  README.md
  init_research_workflow.py
  docs/thesis/
  scripts/
  .claude/settings.local.template.json

Initialize the current project with:

python <research-workflow-kit>/init_research_workflow.py --project .

Add --with-claude-hook when the user wants a Claude Code Stop hook template copied into the project. Use --overwrite only when the user explicitly wants existing project files refreshed from the template.

Ownership rules:

• Global Codex skills in the user's Codex skills directory, usually ~/.codex/skills, define reusable research behavior.
• The cloned research-workflow-kit repository defines reusable project scaffolding.
• Each active project owns its own docs/thesis/ evidence console and scripts/ automation.
• When multiple copies exist, prefer the current project's docs/thesis/ for evidence and the installed skill's scripts/ for reusable skill behavior.

When a project needs persistent organization, suggest this structure:

docs/thesis/
  topic-intake.md
  thesis-brief.md
  project-scope-control.md
  tool-integration-map.md
  task-board-sync.md
  console-file-index.md
  weekly-review.md
  git-version-log.md
  literature-matrix.md
  zotero-literature-hub.md
  zotero-screening-loop.md
  zotero-collection-coverage.md
  paper-readings/
  section-citation-map.md
  section-citation-suggestions.md
  citation-provenance.md
  deep-research-tasks.md
  section-research-packets/
  idea-discovery.md
  experiment-architecture.md
  experiment-runbook.md
  reproducibility-checklist.md
  evidence-promotion-policy.md
  id-lifecycle-policy.md
  benchmark-report-schema.md
  experiment-registry.md
  experiment-integrity-checklist.md
  autoresearch-loop.md
  autoresearch-results.tsv
  autoresearch-state.json
  experiment-reports/
  experiment-notebook-index.md
  claim-evidence-map.md
  material-passport.md
  data-availability.md
  figure-plan.md
  diagram-replica-tasks.md
  diagram-plans/
  network-architecture-figures.md
  spreadsheet-exports.md
  research-materials-index.md
  final-artifact-manifest.md
  final-handoff-verify-report.md
  writing-outline.md
  final-audit.md
  defense-prep.md
  skill-audit-report.md
  workflow-edit-log.md

When the user asks to update thesis status, view thesis progress, resume thesis work, review thesis materials, or decide what to do next, inspect docs/thesis/ first. If the files exist, summarize the current state from the console before recommending next steps. If they do not exist, suggest creating the console, or create it directly when the user asks for implementation.

The core console files are:

File	Primary use
topic-intake.md	advisor title intake, topic decomposition, first research blueprint
thesis-brief.md	Topic, research question, contributions, constraints, timeline
project-scope-control.md	title survival, causal availability, node/structure definition, downgrade/rename policy
tool-integration-map.md	plugin/skill roles by workflow stage
workflow-dashboard.md	one-page current stage, blockers, recent experiments, evidence gaps, audit tier
console-file-index.md	layered source-file map to reduce daily cognitive load
weekly-review.md	short weekly convergence record: stronger/weaker evidence, best experiment, next 1-3 actions
daily-workflow-entry.md	legacy-compatible current research workspace record: focus, blocker, next action, done note
dashboard-web/	React/Vite local web dashboard rendering generated workflow data
workflow-edit-log.md	audit trail for flow-editor writes to Markdown source records
task-board-sync.md	Notion task board and progress sync
git-version-log.md	branch/commit traceability for code and experiments
literature-matrix.md	Literature groups, source status, paper-reading records, citation batches
zotero-screening-loop.md	candidate-paper screening, A/B/C/D labels, Zotero queue, and feedback learning
zotero-collection-coverage.md	Zotero collection coverage by section, claim, and verified citations
paper-readings/	Source-grounded full-paper readers with paper.md, source_map.json, notes, and assets
section-citation-map.md	Section and segment level citation matching
section-citation-suggestions.md	local-only citation candidate ranking for section coverage gaps
citation-provenance.md	citation metadata/support verification, Zotero status, and export trace
deep-research-tasks.md	chapter or section level literature search tasks
section-research-packets/	per-section citation precision packets
idea-discovery.md	Paper pool, idea matrix, novelty risk, shortlist
experiment-architecture.md	experiment code architecture, data/model/train/evaluate boundaries
experiment-runbook.md	run commands, expected outputs, monitoring, failure handling
reproducibility-checklist.md	environment, split, seed, config, artifacts, rerun commands
evidence-promotion-policy.md	ID naming, promotion gates, quick/advisor/final audit tiers
id-lifecycle-policy.md	lifecycle states and relationship rules for SEC/SEG/CLM/EXP/DATA/FIG/MAT/CIT/BMK/ZCOL/DRT/ZREV IDs
evidence-graph.json / evidence-graph.mmd	generated evidence relationship graph
benchmark-report-schema.md	baseline/new metric comparison schema and benchmark guard checklist
experiment-registry.md	Experiment IDs, configs, outputs, metrics, risks
experiment-integrity-checklist.md	leakage, fake ground truth, metric, config, artifact, and scope checks
autoresearch-loop.md	human-supervised iteration plan and verify/guard gate contract
autoresearch-results.tsv	iteration result log
autoresearch-state.json	resumable iteration state
experiment-reports/	baseline comparison and claim-promotion report for each formal run
experiment-notebook-index.md	Reproducible notebook records and console handoffs
claim-evidence-map.md	Claim-to-result-to-figure-to-citation traceability
material-passport.md	identity cards for datasets, experiment outputs, figures, readings, and document artifacts
data-availability.md	dataset provenance, access status, and claim-to-data traceability
figure-plan.md	Figures, tables, captions, input data, status
diagram-replica-tasks.md	Mac draw.io and Windows Visio reference-image replication tasks
diagram-plans/	draw.io and Visio JSON plans for structured diagram reconstruction
network-architecture-figures.md	model structure diagrams, draw.io redraw records, .network.json specs, renderer presets, outputs, QA
spreadsheet-exports.md	export registry for reviewable tables
research-materials-index.md	index of experiment, literature, notebook, figure, and writing materials
final-artifact-manifest.md	stage 11-12 Mac production and Windows compatibility-review manifest for final DOCX/PDF/PPTX/figure/table artifacts
final-handoff-verify-report.md	checksum verification report for the latest final handoff package
writing-outline.md	Chapter goals, evidence, citations, writing status
final-audit.md	Submission, defense, citation, figure, format, and claim audit
defense-prep.md	defense narrative, slide inventory, and Q&A
skill-audit-report.md	generated check of skill structure and outdated assumptions

Workflow

Read the reference files only as needed:

• references/literature-and-citations.md for Semantic Scholar, Scite, Zotero, BibTeX, and pyzotero fallback decisions.
• references/experiment-records.md for notebook and experiment registry workflows.
• references/visualization-rules.md for publication figure rules.
• references/document-production.md for DOCX/PDF/LaTeX production checks.
• references/tool-integration.md for Notion, GitHub, Superpowers, Build Web Data Visualization, CodeRabbit, Vercel, Scite, Zotero, Spreadsheets, draw.io, Presentations, Figma, BioRender, and optional Canva routing.
• references/paper-iteration-loop.md for iterative research-paper execution.
• references/source-map.md for provenance and license notes.

Twelve-Stage Thesis Workflow

Use this stage model when explaining or coordinating the full workflow:

1. Paper planning: $research-paper-plan for Topic Intake -> Research Blueprint when a title is provided; optionally sync tasks to Notion; use idea-discovery.md for paper pool, idea matrix, novelty risk, and shortlist.
2. Literature discovery and review: $semanticscholar-skill, $research-literature-review, $pdf for source-grounded paper readers, Zotero, Scite, long-text citation batching, optional Zotero screening loop, zotero-collection-coverage.md, citation-provenance.md, section-citation-map.md, local section-citation-suggestions.md, and deep-research-tasks.md section packets when a chapter needs tighter citation matching.
3. Experiment question definition: map planned claims to required experiments.
4. Experiment architecture planning: $research-experiment-engineering and $research-code-quality for code boundaries, config-driven entrypoints, and experiment contracts.
5. Research code implementation: Codex coding workflow, Superpowers TDD/debugging, $research-code-quality, GitHub versioning, plugin_gate_advisor.py recommendations, Codex Security for security-sensitive Dashboard/API/remote/CI changes, optional CodeRabbit review before major merges when authenticated.
6. Experiment run and monitoring: $research-experiment-engineering for local_mac CPU-only smoke tests, remote_desktop_4060 primary GPU formal runs, cloud_autodl fallback training with logs, exit code, run summary, artifact recovery, checksums, auto-shutdown policy, and git commit traceability; use $research-autoresearch-loop for human-supervised iterations and verify/guard gates. Use macOS Terminal, VS Code SSH, ssh, scp, or rsync for remote handoff.
7. Experiment recording and result scan: $jupyter-notebook, $research-results-analysis, $research-autoresearch-loop, Data Analytics for data quality and metric diagnostics, Spreadsheets for reviewable exports, and Build Web Data Visualization for result-chart choice and QA; create experiment-reports/EXP-*.md and update benchmark-report-schema.md for formal baseline comparisons; update material-passport.md and data availability when result artifacts become evidence.
8. Results analysis and claim mapping: $research-results-analysis, $research-data-availability, Data Analytics for baseline-delta/anomaly/uncertainty review, Scite for citation-support checks, Build Web Data Visualization for uncertainty chart design, Spreadsheets for claim tables, material-passport.md, benchmark-report-schema.md, citation-provenance.md, and section-citation-map.md.
9. Figure and table planning: $research-paper-figures, with figure-audit-standard.md for Nature-derived claim-first figure QA, Build Web Data Visualization for chart/dashboard/evidence graph design QA, and Product Design for advisor-facing readability and hierarchy review. For route diagrams, method architecture diagrams, mechanism diagrams, workflow diagrams, and system-structure figures, use the PPT-template route first: figure contract -> select academic PPT template -> fill thesis-specific Chinese labels -> inspect thumbnails -> apply PPTX-safe edits -> WPS/PowerPoint manual polish -> SVG/PDF/PNG/PPTX export -> figure audit. Image Gen is optional inspiration only and no longer the default mother draft. Use draw.io when PPT templates are unsuitable or editable vector structure is required. Windows route: reference image -> Visio JSON plan -> .vsdx -> EMF/PDF/PNG export -> copy artifacts back -> audit. Use Python or the Nature-style renderer for data-backed plots, and Spreadsheets for manuscript tables.
10. Paper writing and polishing: split this stage into two gates so writing does not start from premature figure numbering.
    • 10A thesis writing blueprint and first draft: use $research-paper-plan to lock the chapter structure, section titles, writing order, figure/table placement plan, and claim-evidence responsibilities; use $research-results-analysis to turn frozen EXP-103..EXP-106 evidence into section-level result narratives; use $research-paper-writing to draft method, experiment, result analysis, related work, introduction, abstract, and conclusion in that order unless the user supplies another template; use $research-literature-review, $semanticscholar-skill, Zotero, Scite, and $pdf to fill citation gaps through section-citation-map.md, citation-provenance.md, and source-grounded readings; use Spreadsheets for reviewable result/citation/claim tables. Main records: writing-outline.md, claim-evidence-map.md, section-citation-map.md, result-narrative-map.md when present, figure-table-placement.md when present, and nature-style-writing-checklist.md.
    • 10B thesis polish and manuscript integration: after the section draft exists, use $research-paper-writing for section logic, hedging, sentence clarity, unsupported-claim removal, and Chinese thesis tone; use $research-paper-figures to verify captions and figure/text alignment; use Documents for .docx production only after the prose structure is stable; use Zotero/BibTeX export for citation insertion; use $pdf for rendered checks. Final numbering, Word/WPS cross-references, figure captions, and DOCX layout belong here, not before the first full draft.
11. Mac DOCX / PDF / PPTX main production: complete the thesis DOCX main document, PDF export candidate, and PPTX candidate on the Mac first, check evidence/citations/figures, register DOCX/PDF/PPTX/figure/table artifacts in final-artifact-manifest.md, package them with scripts/package_final_handoff.py, and record checksums; use WPS-compatible DOCX, $research-data-availability, Documents plugin for .docx, Pages when useful, Presentations for PPTX, LaTeX doctor first and LaTeX compile only when a TeX runtime is available, and $pdf for rendered checks.
12. Windows compatibility review and final submission preparation: verify the Mac-produced package on the Windows laptop with scripts/verify_final_handoff.py and final-handoff-verify-report.md; open DOCX/PDF/PPTX in Word/WPS/PowerPoint, check layout/export/playback compatibility, final submission details, and defense readiness; use $research-final-audit, $research-data-availability, Presentations, draw.io exports, optional Figma/BioRender visual refinement, optional Canva only when available, Notion task closure, and nature-paper2ppt structure when converting a paper or thesis chapter into a Chinese academic PPTX deck. Final audit must choose quick, advisor, or final tier and check workflow dashboard health, evidence graph gaps, ID lifecycle conflicts, final artifact manifest copied/verified/checksum status, final artifact package verification, material passport completeness, benchmark report schema, figure-audit status, source-grounded reading/citation provenance, Zotero collection coverage, data availability, autoresearch verify/guard status, code contract status, 4060/cloud environment snapshots and remote artifact URI/hash records, and network-architecture draw.io/.network.json plus QA reports.

Output Contract

Always include:

• current project phase
• recommended skill sequence
• required artifacts
• next concrete actions
• risks or blockers
• suggested files to update

If the user asks for full execution, break the work into stages and handle the current stage first. If evidence is missing, name the missing artifact instead of inventing it.