paper-rag

Citing & Filtering ​

/paper-rag:cite generates BibTeX for papers you've already ingested. /paper-rag:mine filters your library's bibliographic metadata — by author, year, journal, keyword, or publication type — without touching the vector store. This guide covers both.

/paper-rag:cite ​

/paper-rag:cite [doc_key|pmid ...] [--all]

Pass one or more doc_keys/PMIDs to cite specific papers, or --all for every paper in the active library that has a metadata.json. The output is raw .bib text, printed verbatim — it's meant to be pasted or redirected into a .bib file as-is, so Claude doesn't reformat or summarize it.

/paper-rag:cite 31978945
bibtex
@article{smith2020adaptive,
  author = {Smith, Jane and Doe, John},
  title = {An adaptive approach to...},
  journal = {Nature Neuroscience},
  year = {2020},
  volume = {23},
  number = {4},
  pages = {512--520},
  doi = {10.1038/s41593-020-00001-1},
  pmid = {31978945},
  pmcid = {PMC7123456}
}

The cite key is generated as {first-author-family}{year}{first-significant-word-of-title} (e.g. smith2020adaptive), lowercased and stripped of non-alphanumerics. If two papers would generate the same key, later ones get a letter suffix (smith2020adaptivea, smith2020adaptiveb, ...).

If a target has no metadata.json — for example a hand-copied local file that never went through a PMID lookup — the script warns on stderr and skips it; Claude relays that warning after the BibTeX output rather than silently dropping the paper.

/paper-rag:ask calls /paper-rag:cite automatically to build its ## References section — see Asking Questions.

/paper-rag:mine ​

/paper-rag:mine is metadata mining, not semantic search: it scans every ingested paper's metadata.json and filters by bibliographic fields. Use it to answer questions like "what did I publish as first author since 2020?" — questions about which papers exist, not about what's in them.

/paper-rag:mine [--author NAME] [--as first|last|any] [--year Y|Y-Y] [--journal J]
    [--keyword K] [--pub-type T]
  • --author NAME — case-insensitive substring match against each paper's author family field. If omitted, falls back to author_name from config.json (set via /paper-rag:whoami, see Tags & Homes); if that isn't set either, every paper is returned unfiltered.
  • --as first|last|any (default any) — restrict the author match to first-author, last-author, or any position in the author list.
  • --year Y or --year Y1-Y2 — a single year or an inclusive range.
  • --journal J — case-insensitive substring match against the journal name.
  • --keyword K — case-insensitive substring match against any of the paper's keywords.
  • --pub-type T — case-insensitive substring match against any of the paper's publication types (e.g. Review, Journal Article).

All filters combine with AND. Example — your own last-author papers from 2019–2023 in a specific journal:

/paper-rag:mine --as last --year 2019-2023 --journal "Nature Neuroscience"

Each result includes position — first/last/middle/none, relative to the matched author — whenever an author filter was applied (omitted otherwise). Claude presents results as a table (title, journal, year, and author position when relevant) rather than dumping raw JSON, and says plainly when nothing matched instead of returning an empty table silently.