Install any skill in seconds. Free to start, no credit card required.
Get Started Free →This skill provides reference guidance for citation verification in academic writing. Use when the user asks about "citation verification best practices", "how to verify references", "preventing fake citations", or needs guidance on citation accuracy. This skill supports ml-paper-writing by providing detailed verification principles and common error patterns.
.claude/skills/galaxy-dawn-citation-verification/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 62% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 11% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 27% | 0% |
| case-05 | ✓→✓ | = Same ✓ | 116% | 0% |
| case-01 | ✓→✓ | = Same ✓ | 72% | 0% |
A reference guide for citation verification in academic paper writing, providing verification principles and best practices.
Core Principle: Proactively verify every citation during the writing process using programmatic or canonical scholarly sources first: arXiv, DOI/CrossRef, Semantic Scholar, publisher landing pages, and Zotero metadata. Google Scholar is useful for manual discovery, but it is not the canonical verification authority.
Citation issues in academic papers seriously impact research integrity:
These issues can lead to:
Special risk with AI-assisted writing: AI-generated citations have approximately 40% error rate; every citation must be verified via WebSearch.
This skill provides verification principles based on canonical scholarly metadata and claim-level checking:
Core idea: Verify immediately when adding a citation, rather than checking after writing is complete.
Preferred authority order:
Verification steps:
Information that must match:
Key principle: When citing a specific claim, you must confirm the claim actually appears in the paper.
Need a citation during writing
↓
Find DOI / arXiv ID / publisher page / verified Zotero item
↓
Verify metadata with CrossRef / arXiv / Semantic Scholar / publisher / Zotero
↓
Confirm paper details
↓
Get BibTeX
↓
(If citing a specific claim) Verify the claim
↓
Add to bibliographyKey point: Verification is part of the writing process, not a separate post-processing step.
The verification principles of this skill are integrated into the Citation Workflow of the ml-paper-writing skill.
Auto-trigger: Citation verification is automatically executed when writing papers with the ml-paper-writing skill.
Manual reference: Refer to this skill when you need detailed verification principles.
Scenario: Need to cite the Transformer paper
Step 1: WebSearch lookup
Query: "Attention is All You Need Vaswani 2017"
Result: Found multiple sources for the paper
Step 2: Google Scholar verification
Query: "site:scholar.google.com Attention is All You Need Vaswani"
Result: ✅ Paper exists, 50,000+ citations, NeurIPS 2017
Step 3: Confirm details
- Title: "Attention is All You Need"
- Authors: Vaswani, Ashish; Shazeer, Noam; Parmar, Niki; ...
- Year: 2017
- Venue: NeurIPS (NIPS)
Step 4: Get BibTeX
- Click "Cite" on Google Scholar
- Select BibTeX format
- Copy BibTeX entry
Step 5: Add to bibliography
- Paste into .bib file
- Use \cite{vaswani2017attention} in the paperIf the paper cannot be verified through canonical sources:
[CITATION NEEDED] markerIf information doesn't match:
[CITATION NEEDED] to mark explicitly❌ Wrong approach:
✅ Correct approach:
Core Principle: Proactively verify every citation during the writing process using WebSearch and Google Scholar.
Key Steps:
Failure handling: When verification fails, mark as [CITATION NEEDED] and clearly notify the user.
Integration: The principles of this skill are integrated into the ml-paper-writing skill for automatic verification.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | pass→pass | 8,517 | 6,768 | -21% | 1 | 1 | 0% | 1,239 | 2,671 | +116% | 0 | 0 | — |
case-01 | pass→pass | 13,302 | 12,252 | -8% | 1 | 1 | 0% | 2,139 | 3,673 | +72% | 0 | 0 | — |
case-02 | pass→pass | 9,329 | 5,030 | -46% | 1 | 1 | 0% | 1,572 | 2,500 | +59% | 0 | 0 | — |
case-03 | pass→pass | 19,273 | 3,177 | -84% | 1 | 1 | 0% | 1,755 | 2,206 | +26% | 0 | 0 | — |
case-04 | fail→pass | 8,515 | 3,371 | -60% | 1 | 1 | 0% | 1,431 | 2,318 | +62% | 0 | 0 | — |
case-06 | fail→pass | 13,899 | 6,372 | -54% | 1 | 1 | 0% | 2,355 | 2,612 | +11% | 0 | 0 | — |
case-07 | fail→pass | 16,282 | 38,927 | +139% | 1 | 1 | 0% | 2,840 | 3,619 | +27% | 0 | 0 | — |
case-08 | pass→pass | 8,249 | 8,672 | +5% | 1 | 1 | 0% | 1,453 | 3,039 | +109% | 0 | 0 | — |
case-09 | pass→pass | 14,206 | 10,536 | -26% | 1 | 1 | 0% | 2,240 | 3,240 | +45% | 0 | 0 | — |
case-10 | pass→pass | 8,783 | 3,091 | -65% | 1 | 1 | 0% | 1,379 | 2,065 | +50% | 0 | 0 | — |
case-11 | pass→pass | 9,169 | 3,034 | -67% | 1 | 1 | 0% | 1,501 | 2,163 | +44% | 0 | 0 | — |
case-12 | pass→pass | 4,338 | 4,756 | +10% | 1 | 1 | 0% | 615 | 2,449 | +298% | 0 | 0 | — |
case-13 | pass→pass | 13,776 | 3,336 | -76% | 1 | 1 | 0% | 2,127 | 2,186 | +3% | 0 | 0 | — |
case-14 | pass→pass | 9,869 | 6,163 | -38% | 1 | 1 | 0% | 1,471 | 2,645 | +80% | 0 | 0 | — |
case-15 | fail→fail | 4,721 | 3,658 | -23% | 1 | 1 | 0% | 818 | 2,294 | +180% | 0 | 0 | — |
case-16 | pass→pass | 11,914 | 8,828 | -26% | 1 | 1 | 0% | 1,935 | 3,121 | +61% | 0 | 0 | — |
case-17 | pass→pass | 12,893 | 6,023 | -53% | 1 | 1 | 0% | 2,071 | 2,601 | +26% | 0 | 0 | — |
case-18 | pass→pass | 31,343 | 4,697 | -85% | 1 | 1 | 0% | 2,292 | 2,425 | +6% | 0 | 0 | — |
case-19 | pass→pass | 14,302 | 10,967 | -23% | 1 | 1 | 0% | 2,473 | 3,587 | +45% | 0 | 0 | — |
case-20 | pass→pass | 5,030 | 5,205 | +3% | 1 | 1 | 0% | 884 | 2,498 | +183% | 0 | 0 | — |
case-21 | pass→pass | 8,423 | 8,799 | +4% | 1 | 1 | 0% | 1,575 | 3,137 | +99% | 0 | 0 | — |
case-22 | pass→pass | 13,176 | 15,072 | +14% | 1 | 1 | 0% | 2,402 | 4,496 | +87% | 0 | 0 | — |
case-23 | pass→pass | 12,555 | 13,876 | +11% | 1 | 1 | 0% | 2,182 | 3,793 | +74% | 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.