| name | gpt-pro-paper-brainstormer |
| description | Use GPT Pro for research-paper brainstorming, claim sharpening, novelty analysis, related-work positioning, reviewer objections, and experiment story design. Best for algorithm/research ideas where Codex alone is too implementation-biased. |
GPT Pro Paper Brainstormer
Gather only the evidence needed to test the research claim, then ask GPT Pro to act as a skeptical research advisor.
Workflow
- Clarify the raw idea and target venue level if known.
- Use
$bundle-algorithm-context if code, experiments, docs, or logs matter.
- Read references/paper_brainstorm_prompt.md, then use
$gpt-pro-question-window to ask GPT Pro.
- Capture the raw response and record a separate Codex verdict on the same
bridge-thread-id.
- Codex synthesizes:
- Strongest paper claim.
- Weakest/most vulnerable claim.
- Required experiments.
- Whether this is paper-worthy, workshop-worthy, or just engineering.
Completion criterion: claims and related-work references are marked as verified or unverified, the required experiments are concrete, and the publishability judgment is preserved separately from Codex's local verdict.
Prompt dimensions
Ask GPT Pro to cover:
- One-sentence paper claim.
- Problem pain and why now.
- Novelty type: algorithm, data, benchmark, system, analysis, application.
- Nearest likely related work clusters.
- What is actually new vs rebranding.
- Strongest title/abstract angle.
- Experimental story.
- Killer baseline and ablations.
- Reviewer objections.
- Minimal evidence needed to make the claim believable.
- A brutally honest publishability judgment.
Codex-side caution
GPT Pro may invent related work or overstate novelty. After GPT Pro responds, Codex should mark related-work items as unverified unless it has searched or the user supplied citations.