Top-level research workflow that orchestrates the three-stage process: idea discovery, experiment pipeline, and paper writing. Use when the user wants the complete end-to-end flow.
Run the full research workflow for: $ARGUMENTS
This is the top-level orchestrator for the complete research lifecycle.
It coordinates three stage skills:
/idea-discovery -> /experiment-pipeline -> /paper-writing
Use this skill when the user wants the full end-to-end workflow from:
broad direction -> validated idea -> completed experiments -> compiled paper
Do not use this skill when the user clearly wants only one stage. In that case, invoke the specific stage skill directly.
RESEARCH_PIPELINE_REPORT.md - Final top-level summary artifactopus - Use for problem framing, novelty arbitration, experiment sufficiency judgments, and final go/no-go decisions.sonnet - Use for drafting, coding, plotting, LaTeX edits, and other mechanical execution tasks.gpt-5.4 - Use for independent critical review and paper-quality scoring.gemini-3-pro-preview - Use inside /paper-illustration for layout and style optimization.gemini-3-pro-image-preview - Use inside /paper-illustration for final non-data figures.If the host cannot select opus or sonnet directly, preserve the same role split with the strongest available local replacement:
gpt-5.4| Role | Default Model | Responsibility | Persona |
|---|---|---|---|
| Strategy Lead / Final Arbiter | opus | Final framing, novelty arbitration, evidence thresholding, major tradeoffs | Senior professor in multi-agent systems, formation control, cooperative robotics, and distributed control; long-time reviewer for IEEE field journals such as TAC, T-RO, T-ASE, T-CYB, and T-SMC |
| Execution Lead | sonnet | Writing, code changes, figure script generation, experiment plumbing, LaTeX edits | Strong research engineer and technical writer who executes precisely against the chosen plan |
| External Reviewer | gpt-5.4 | Independent criticism, acceptance risk assessment, missing evidence detection | Skeptical IEEE journal reviewer with deep experience in multi-agent formation control, control theory, and experimental validation |
| Illustration Stack | Gemini + Codex | Layout, rendering, and strict visual QA for non-data figures | Publication-figure specialist for IEEE control and robotics papers |
Invoke:
/idea-discovery "$ARGUMENTS"
Expected outputs:
IDEA_REPORT.mdFINAL_PROPOSAL.mdNOVELTY_CHECK.mdDo not proceed to Stage 2 unless there is:
If the idea stage fails, stop here and report the best salvage path.
Invoke:
/experiment-pipeline "FINAL_PROPOSAL.md"
Expected outputs:
EXPERIMENT_PLAN.mdRESULTS_SUMMARY.mdNARRATIVE_REPORT.mdDo not proceed to Stage 3 unless there is:
If experiments are incomplete or unsupported claims dominate, stop here and report what is missing.
Invoke:
/paper-writing "NARRATIVE_REPORT.md"
Expected outputs:
PAPER_PLAN.mdfigures/paper/paper/main.pdfThe pipeline ends only when the manuscript compiles or a clear blocking reason is reported.
Start with /idea-discovery.
Checkpoint summary should include:
The Strategy Lead owns the accept / reject / re-scope decision at this gate.
If HUMAN_CHECKPOINT=true, pause after presenting the summary.
Pass the chosen proposal into /experiment-pipeline.
Checkpoint summary should include:
The Strategy Lead decides whether the evidence is sufficient to enter the paper stage. The Execution Lead owns implementation and result packaging.
If HUMAN_CHECKPOINT=true, pause after presenting the summary.
Pass the narrative package into /paper-writing.
Checkpoint summary should include:
The Execution Lead handles drafting and figure generation. The External Reviewer and Strategy Lead handle final quality judgment.
Write RESEARCH_PIPELINE_REPORT.md in this structure:
# Research Pipeline Report
## Executive Summary
- Topic:
- Final status:
- Recommended next action:
## Stage 1: Idea Discovery
- Recommended idea:
- Novelty status:
- Main risks:
- Output files:
## Stage 2: Experiment Pipeline
- Main completed runs:
- Main findings:
- Unsupported claims:
- Output files:
## Stage 3: Paper Writing
- Paper status:
- Figure status:
- Compile status:
- Output files:
## Remaining Gaps
- Item:
- Why it matters:
## Final Deliverables
- `FINAL_PROPOSAL.md`
- `RESULTS_SUMMARY.md`
- `NARRATIVE_REPORT.md`
- `paper/main.pdf`
Stop and report clearly if:
Do not invent continuity when a stage is blocked. Report the exact missing artifact or decision.
This orchestrator should use only this sequence:
/idea-discovery
/experiment-pipeline
/paper-writing
No other top-level stage commands are part of this orchestrator.