paper-search

Paper Search

When To Use

• The user asks to search papers for a research direction
• The user wants to restrict the time range, result count, or data sources
• The user wants OpenReview or arXiv results prioritized
• The user needs a structured paper list for later reading, filtering, or summarization
• The user wants to see sample papers, experiment scales, or writing patterns of a direction in target journals (e.g. Q3/Q4 venues)

Workflow Integration

• In a full paper project, research-project-workflow first coordinates the directory layout and stage order; then this skill searches candidate papers.
• This skill only returns a candidate paper list; PDF download, text conversion, factual paper cards, experiments, and manuscript writing are outside its scope.
• In a paper project, organize or save the script JSON output to ref_papers/source.json under the exact cwd; if the dispatch explicitly supplies an existing user layout, follow that layout.
• For finally selected arXiv papers, use the arxiv skill to verify ID, version, BibTeX, and citation information.
• If public PDFs must be fetched and converted to Markdown, the Search agent uses the pdf-to-markdown skill; the orchestration layer never performs conversion directly.
• After selected papers are verified and converted, Search uses paper-material-package to create ref_papers/paper-notes.md. Do not add summaries or survey prose to this candidate-search Skill.

Inputs

Extract from the user request:

• Search direction / topic (required)
• Start date since (optional, format YYYY-MM-DD; default 2026-01-01)
• End date until (optional, format YYYY-MM-DD; default today)
• Result count max-results (optional, default 10)
• Data sources sources (optional; default searches both openreview and arxiv)

By default only query is needed; the remaining parameters use their defaults. Add a time range, data sources, or result count only when the user explicitly asks.