跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
搜索与研究

ref-check

Careful bibliography verification and entry-addition for papers — verify every author, title, journal, volume, pages, year, arXiv ID, and DOI against authoritative online sources (INSPIRE, arXiv, CrossRef/DOI, zbMATH, publisher) before anything enters or stays in a .bib file. Use whenever adding bib entries, auditing a bibliography, checking citations, fixing references, or when asked to "check the references". Encodes the report-then-fix pattern for full-file audits and the folklore-error traps (wrong years, truncated author lists, mislabeled keys).

BootLoops-ai · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add BootLoops-ai/skills --skill ref-check

技能说明


name: ref-check description: Careful bibliography verification and entry-addition for papers — verify every author, title, journal, volume, pages, year, arXiv ID, and DOI against authoritative online sources (INSPIRE, arXiv, CrossRef/DOI, zbMATH, publisher) before anything enters or stays in a .bib file. Use whenever adding bib entries, auditing a bibliography, checking citations, fixing references, or when asked to "check the references". Encodes the report-then-fix pattern for full-file audits and the folklore-error traps (wrong years, truncated author lists, mislabeled keys).

ref-check — careful reference verification

The iron rule, and why it exists

NEVER guess or trust-from-memory an arXiv ID, journal reference, year, author list, page range, or title. Every factual field is verified against a fetched authoritative record before an entry is added, approved, or "fixed." Training-data recall of bibliographic details is a hypothesis to verify, never a source. If a field cannot be verified after trying at least two sources, it is FLAGGED, not silently kept or invented.

The rule is this severe because bibliographies are not copied from sources — they are copied from other bibliographies. An error, once printed, outlives its origin: the next hundred authors copy the entry, not the paper, and the error acquires the authority of repetition. A language model trained on that literature has learned the majority text, which for a propagated error IS the error — model recall of a reference reproduces the folklore version with full confidence. This is why "it looks right" is worthless here, and why an entry that half the field cites one way can still be wrong. Verification means fetching the record closest to the publisher and comparing field by field; nothing else counts.

A second consequence: an entry that a model drafted from memory is not a degraded citation — it is a fabrication with correct formatting. The observed pattern (see traps below) is a real identifier carrying invented co-authors, a paraphrased title, or another paper's journal block. Plausibility is the failure mode, not the reassurance.

Authoritative sources, by literature

The universal spine is CrossRef / the publisher's DOI landing page — canonical journal, volume, pages, year for anything with a DOI. Around it, use the registry native to the literature:

  1. INSPIRE-HEP (physics/hep) — fetch BibTeX directly: https://inspirehep.net/api/literature?sort=mostrecent&size=1&q=arxiv%3A<ID>&format=bibtex (URL-encode : as %3A; old-style IDs URL-encode the slash: q=arxiv%3Ahep-ph%2F<YYMMNNN>). Cross-check the arXiv abs page.
  2. arXiv abs page (https://arxiv.org/abs/<ID>) — title, full author list, journal-ref line. v1 titles sometimes differ from the published title; the published title wins when the entry cites the journal.
  3. CrossRef / publisher DOI page — the arbiter for journal fields.
  4. zbMATH / MathSciNet / Project Euclid / numdam — mathematics.
  5. ADS — astronomy; PubMed/PMC — life sciences; DBLP — computer science (conference/proceedings metadata that CrossRef often garbles).
  6. Zenodo / the software's own CITATION file — software and datasets.
  7. Gallica / archive.org / WorldCat / publisher record — classical books and pre-DOI literature.

Search aggregators (Google Scholar, Semantic Scholar) are for FINDING a work, never for verifying it. They merge records, inherit upstream errors, and invent metadata for scraped PDFs. Once found, verify against the native registry or publisher.

Sleep ~2s between external calls; try a second source before declaring anything unverifiable. When two authoritative sources disagree, the one closest to the publisher wins for journal fields, and the disagreement itself goes in the audit log — it usually marks a folklore error in the losing source.

Field-by-field comparison (semantic, not textual)

  • Authors — every author present, correct order, correct spelling including diacritics. TeX escapes vs unicode ({\'e} = é) are equal. A missing, extra, or misspelled author = FIX. Watch homonyms: same-surname collaborators are routinely merged into one person, and given names swapped between them.
  • Title — word-level. Brace/capitalization/TeX-markup differences = formatting. Wrong, missing, or extra words = FIX.
  • Journal, volume, number, pages, year — exact. Standard abbreviation vs full journal name = MINOR. Wrong volume/pages/year = FIX. Translated-journal pairs (Russ. Math. Surv./Uspekhi Mat. Nauk; Sov. Phys. JETP/ZhETF): either is fine if internally consistent; a mismatched year between the pair = FIX.
  • arXiv ID, DOI — exact, and the DOI must resolve to the right work: fetch the landing page and compare its title. A syntactically valid DOI pointing at an erratum, a comment, or a different paper by the same group is a FIX that no string comparison catches.
  • Entry type and roles — chapters in collections credit the chapter authors, with editors as editors; proceedings carry the conference year vs publication year distinction explicitly.

Classify each entry: OK (every factual field compared against a fetched record and matching) · MINOR (formatting only, no action) · FIX (any factual discrepancy — record field: ours -> authoritative + source URL) · UNVERIFIABLE (state exactly what was tried) · INTERNAL (self-references/companion notes — list, exempt).

Never mark OK without having actually fetched and compared. "Looks right" is not a verification.

Folklore traps (all observed in practice)

Each of these is a measured failure mode from real bibliography audits, not a hypothetical. They are the classes that survive casual checking because the entry looks healthy.

  • Wrong years that circulate — a result becomes attached to the wrong year and the whole literature repeats it: e.g. a landmark irrationa
阅读完整来源 ↗