Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Guide to JabRef open-source BibTeX and BibLaTeX reference manager
.claude/skills/brycewang-stanford-jabref-reference-guide/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 28% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 50% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 65% | 0% |
| case-02 | ✓→✓ | = Same ✓ | 30% | 0% |
| case-03 | ✓→✓ | = Same ✓ | 48% | 0% |
JabRef is an open-source, cross-platform reference manager with over 4,000 GitHub stars, specifically designed for researchers who work with BibTeX and BibLaTeX bibliographies. Written in Java, it runs on Windows, macOS, and Linux, providing a graphical interface for managing bibliographic databases that integrates seamlessly with LaTeX-based writing workflows.
Unlike general-purpose reference managers that treat BibTeX export as an afterthought, JabRef is built from the ground up around the BibTeX format. Every feature, from the entry editor to the search system, understands BibTeX entry types and field semantics natively. This means researchers who write primarily in LaTeX get a tool that speaks their language without the impedance mismatch of converting between proprietary formats.
JabRef supports both BibTeX and BibLaTeX formats, handles cross-references, string constants, and preambles, and provides powerful tools for maintaining bibliography quality. It can fetch metadata from online databases, check entries for completeness, detect duplicates, and generate citation keys following customizable patterns. For researchers who prefer to keep their bibliographic data in plain-text BibTeX files under version control, JabRef is the reference manager of choice.
JabRef can be installed through multiple methods:
Direct Download:
Package Managers:
brew install --cask jabrefsudo snap install jabrefflatpak install org.jabref.jabrefwinget install JabRef.JabRefInitial Configuration:
[auth][year] produces keys like Smith2024[auth:lower][year][veryshorttitle:lower] produces smith2024deepEntry Management: JabRef provides a table view and an entry editor for managing bibliography entries. The entry editor shows all fields relevant to the selected entry type (article, book, inproceedings, etc.) and validates field content against BibTeX specifications.
Online Database Search: Search and import references from multiple academic databases directly within JabRef:
Duplicate Detection: JabRef identifies potential duplicate entries using configurable similarity algorithms that compare titles, authors, years, and other fields. The merge dialog allows you to combine duplicates while preserving the most complete metadata from each entry.
Group Management: Organize entries into hierarchical groups based on:
Quality Checks: Run integrity checks on your database to identify:
Push to Editor: Send citation commands directly to your LaTeX editor. JabRef supports integration with:
Setting Up a Project Bibliography:
.bib file in your LaTeX project directory.bib file\bibliography{references} or \addbibresource{references.bib}Collaborative Writing:
.bib file in a Git repository alongside your LaTeX sourceIntegration with Overleaf:
.bib fileJournal Submission Preparation:
Custom Entry Types: Define custom BibTeX entry types for specialized references not covered by the standard types. This is useful for datasets, software, standards documents, and other non-traditional academic sources.
Journal Abbreviation Management: JabRef includes a comprehensive database of journal name abbreviations. Configure automatic abbreviation or expansion of journal names to match specific publisher requirements.
String Constants: Use BibTeX string constants to define frequently used values (journal names, publisher details, conference series names) once and reference them throughout your database. This ensures consistency and simplifies bulk updates.
Web Search Customization: Configure custom web search providers or API endpoints for specialized databases relevant to your field.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 47,108 | 44,658 | -5% | 1 | 1 | 0% | 3,005 | 3,848 | +28% | 0 | 0 | — |
case-02 | pass→pass | 10,129 | 4,898 | -52% | 1 | 1 | 0% | 1,845 | 2,398 | +30% | 0 | 0 | — |
case-03 | pass→pass | 13,613 | 17,377 | +28% | 1 | 1 | 0% | 2,403 | 3,557 | +48% | 0 | 0 | — |
case-04 | fail→pass | 12,441 | 37,922 | +205% | 1 | 1 | 0% | 1,905 | 2,861 | +50% | 0 | 0 | — |
case-05 | fail→fail | 13,156 | 10,222 | -22% | 1 | 1 | 0% | 2,094 | 3,153 | +51% | 0 | 0 | — |
case-06 | pass→pass | 10,580 | 8,290 | -22% | 1 | 1 | 0% | 1,649 | 2,975 | +80% | 0 | 0 | — |
case-07 | fail→fail | 17,244 | 16,159 | -6% | 1 | 1 | 0% | 2,390 | 4,012 | +68% | 0 | 0 | — |
case-08 | pass→pass | 10,053 | 9,478 | -6% | 1 | 1 | 0% | 1,688 | 2,981 | +77% | 0 | 0 | — |
case-09 | pass→pass | 6,150 | 8,119 | +32% | 1 | 1 | 0% | 1,063 | 2,572 | +142% | 0 | 0 | — |
case-10 | pass→pass | 12,275 | 14,873 | +21% | 1 | 1 | 0% | 2,066 | 3,496 | +69% | 0 | 0 | — |
case-11 | fail→pass | 14,867 | 19,152 | +29% | 1 | 1 | 0% | 2,511 | 4,136 | +65% | 0 | 0 | — |
case-12 | pass→pass | 12,597 | 12,550 | -0% | 1 | 1 | 0% | 2,171 | 3,529 | +63% | 0 | 0 | — |
case-13 | pass→pass | 12,831 | 6,016 | -53% | 1 | 1 | 0% | 1,875 | 2,495 | +33% | 0 | 0 | — |
case-14 | pass→pass | 14,685 | 18,416 | +25% | 1 | 1 | 0% | 2,569 | 3,977 | +55% | 0 | 0 | — |
case-15 | pass→pass | 14,193 | 11,499 | -19% | 1 | 1 | 0% | 2,297 | 3,506 | +53% | 0 | 0 | — |
case-16 | pass→pass | 14,209 | 15,504 | +9% | 1 | 1 | 0% | 2,464 | 3,644 | +48% | 0 | 0 | — |
case-17 | fail→fail | 4,715 | 7,567 | +60% | 1 | 1 | 0% | 821 | 2,429 | +196% | 0 | 0 | — |
case-18 | pass→pass | 10,301 | 11,244 | +9% | 1 | 1 | 0% | 1,717 | 3,087 | +80% | 0 | 0 | — |
case-19 | pass→pass | 16,077 | 14,812 | -8% | 1 | 1 | 0% | 2,205 | 3,805 | +73% | 0 | 0 | — |
case-20 | pass→pass | 7,110 | 4,437 | -38% | 1 | 1 | 0% | 1,060 | 2,227 | +110% | 0 | 0 | — |
case-21 | pass→pass | 11,042 | 9,773 | -11% | 1 | 1 | 0% | 1,886 | 3,023 | +60% | 0 | 0 | — |
case-22 | pass→pass | 11,166 | 11,896 | +7% | 1 | 1 | 0% | 2,107 | 3,447 | +64% | 0 | 0 | — |
case-23 | pass→pass | 8,044 | 9,840 | +22% | 1 | 1 | 0% | 1,461 | 3,148 | +115% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted. The headline lift of +13 percentage points is the difference between those two pass rates over the 23 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.