Paper & Solution Research Skill

Steps

1. Determine Search Query:

   • Use $ARGUMENTS if provided
   • Otherwise extract related keywords from the competition name in KAGGLE_DIRECTION.md

2. Research via WebSearch (search the following in parallel):

   • Related papers on arXiv
   • Top solutions from similar past competitions across platforms (Kaggle / grand-challenge.org / MICCAI Challenges / CodaBench — not limited to Kaggle)
   • Related methods on Papers With Code
   • Related implementations on GitHub

3. Summarize each paper/solution:

   • Title, authors, URL
   • Method summary (3-5 lines)
   • Model/architecture used
   • Dataset & evaluation metrics
   • Key results
   • Applicability to this competition

4. Return the full research output:

   • This skill runs in context: fork (read-only Explore agent), so do not save files inside the fork
   • Return the full research output, organized by category (dataset papers, method papers, benchmark papers, tools/libraries), as the final response
   • The receiving main conversation appends and saves it to survey/papers/maybe_related_research.md

5. Propose transfer ideas:

   • List methods potentially applicable to this competition
   • Evaluate implementation difficulty and expected impact