finding-papers

name	finding-papers
description	Search academic literature, manage the search queue, download papers, and verify sources. Use when the user wants to find papers, when working search-queue.md, or when adding references.

Finding Papers

Which search tool to use

• Built-in make commands for academic papers. They query dedicated databases (200M+ papers) and return structured results — citation counts, DOIs, years, BibTeX. Always try these first for paper discovery.
• Web search for everything else — a researcher's lab page, whether a tool exists, non-academic facts. Don't substitute web search for the make commands: it returns blog posts and news mixed with papers.
• If a make command fails, fall back to web search, tell the user what happened, and suggest make check or ./setup.sh.

Commands

• make search-py QUERY="topic" — Semantic Scholar, human-readable results with citation counts and years (use this for presenting options to the user)
• make search QUERY="topic" — Semantic Scholar, BibTeX output (use this when appending an entry to references.bib)
• make search-openalex QUERY="topic" — OpenAlex (250M+ papers, alternative source)
• make search-author AUTHOR="Jane Smith" — papers by author, with h-index
• make fetch-arxiv ID="2301.00001" / make fetch-doi DOI="10.1038/..." — download to sources/
• make add-paper DOI="..." — add to papis library; make bib — regenerate references.bib
• make verify DOI="..." — check citation support/contradiction (scite)
• make identify-doi PDF="sources/paper.pdf" — find a DOI inside a PDF

The search queue (search-queue.md)

The project's literature backlog: every paper still needed, in three tiers.

• P0 — must have before drafting. Positioning papers (closest neighbors to cite and differentiate from), core framing sources, evidence for the most vulnerable claims.
• P1 — search while drafting, when a section reveals a real gap.
• P2 — only if needed later. Don't search these until a draft demands it.

Add to the queue after outline changes, after a paper points somewhere worth following, when the user names a gap, or when a section draft's "Gaps" section reveals one. Always assign a tier.

Work the queue when the user says "find papers" (check the queue before asking what to search), and always P0 before P1 before P2:

1. Pick the highest-priority open search
2. Run it with make search-py or make search-openalex using the listed keywords
3. No good results → try alternative keywords or the other tool
4. Update the entry: results, status ([x] found, [-] failed + what was tried)
5. Move completed searches to "Sources Already Collected" (or archive.md)

Good-enough-to-draft threshold: the queue defines a gate — typically all positioning papers read plus one strong source per evidence-dependent section. Check it after each search batch. Don't let research become an excuse not to write.

When the user asks for papers on a topic

1. make search-py QUERY="the topic"
2. Present readable results: title, year, citation count
3. Ask which to download; fetch with make fetch-doi / make fetch-arxiv
4. make add-paper DOI="..." to add to the library
5. Automatically create a reading note (see the reading-notes skill) — don't wait to be asked

Adding a citation to the manuscript

1. Check bibliography/references.bib for the key
2. Missing → make search QUERY="paper title" and append the BibTeX, or make add-paper DOI="..." then make bib
3. Insert [@key] at the right place — keys must exist in references.bib or the PDF build breaks