Jarvis Research Idea Generator Pro

Problem

When generating research topics, the hard part is not collecting papers, but turning fragmented literature into executable research questions: which areas have dense evidence, which are only trend-driven clusters, and which open problems can support testable ideas. This skill is for research ideation. It takes a research domain and optional seed ideas, then produces evidence-linked, evaluable research questions and candidate ideas.

How It Works

The skill runs an auditable ideation pipeline:
- Retrieval and filtering: collects recent top-venue papers from sources such as OpenReview, PMLR, NeurIPS, ACL Anthology, CVF, ACM, and AAAI, while excluding workshops, demos, posters, and irrelevant papers.
- Domain tree construction: maps papers into hierarchical nodes with shared assumptions, limitations, and open-problem hints instead of producing a flat summary.
- Multi-agent brainstorming: uses role-specific agents for detail mining, cross-domain transfer, and divergent thinking to propose, critique, and synthesize candidate questions, then ranks the top questions.
- Idea generation and evaluation: for each top question, generates candidate ideas, runs targeted novelty checks, scores them against a rubric, and rejects, retries, or replaces weak ideas.
- Process traceability: writes intermediate artifacts under outputs/<run-date>/, including paper cards, retrieval reports, novelty checks, and process_record.md.

Boundaries

It is suitable for topic selection, literature triangulation, and drafting testable research plans, but it does not claim exhaustive coverage. If recent evidence is sparse, the skill stops full analysis or records a fallback. If novelty evidence is insufficient, it should not be treated as proof of novelty. Provide a sufficiently specific domain, because overly broad inputs can weaken the paper corpus and domain tree.