SmartLib Citation Authenticity Checker

The Specific Problem It Solves

In academic writing, the authenticity and accuracy of references are cornerstones of a paper's credibility. A new challenge has emerged with AI writing tools: when users generate references with models like ChatGPT or Wenxin Yiyan, they may inadvertently create non-existent papers, incorrect authors, or wrong journal names—known as AI hallucinations. Additionally, researchers face inefficiency and errors when manually checking references during draft review or batch editorial checks.

This skill addresses two core pain points:
* AI Hallucination Verification: Rapidly verifying if AI-generated references actually exist in academic databases.
* Citation Accuracy Auditing: Performing automated, batch comparison and correction of references from manuscripts or pasted lists.

How the Skill Works

Built on global-biblio-base's literature search capabilities, this skill implements citation checking and correction through five key steps.

Core Capabilities

• Multi-Format Parsing: Automatically identifies and parses various citation formats including GB/T 7714, APA, MLA, Chicago, and BibTeX, with support for batch import from .bib files.
• Online Authenticity Verification: Connects via the SmartLib API to a global database of over 1.2 billion documents (including 300+ sources like Scopus, WoS, EI, PubMed) to verify each reference's existence—a critical defense against AI hallucinations.
• Intelligent Field Comparison & Correction: Compares original citations field-by-field (title, authors, year, journal, volume/issue, pages, DOI) with database records, calculates match scores, and provides corrected versions for errors (using [DELETED]/[ADDED] markers).
• Structured Report Generation: Outputs a complete HTML report featuring difference markers, clickable verification links, statistical analysis charts, and one-click download of corrected reference lists.

Key Workflow

1. Parse Citations: Extract core fields like title, authors, and year from each reference in the user input.
2. Online Verification: Search via the SmartLib API using an optimized strategy: parallel searches (8 items per batch) and intelligent prioritization (starting with title keywords + year, falling back to relaxed conditions on failure). Each API call requires obtaining a single-use consume_token (60-second TTL) via /consume for security.
3. Match & Compare: Compute an overall match score based on title similarity (60%), author match (25%), and year match (15%). Classify results as VERIFIED, MISMATCH, NOT_FOUND, or FUZZY_MATCH.
4. Diff Analysis: For MISMATCH items, perform detailed field-level comparison and provide specific correction suggestions.
5. Generate Report: Create a citation_check_report.html file, automatically opened in preview. The report includes original citations alongside old/new national standards (GB/T 7714-2015/2025), multi-format outputs (APA, MLA, etc.), and statistical analyses like bar charts.

Capability Boundaries & Important Notes

Applicable Scenarios

• Core Scenarios: Verifying AI-generated references; self-checking citations before paper submission; batch validating pasted reference lists.
• Extended Scenarios: Processing .bib files exported from Zotero/EndNote; conducting citation statistical analysis (e.g., publication year distribution).

Key Limitations & Considerations

• Quota & Billing: This skill shares quota and wallet with global-biblio-base. Quota is consumed per actual API call (5 interfaces, each call = 1 quota), not per verification session. When quota is exhausted (status code 429), service is denied until the plan is upgraded.
• Security Mechanism: A mandatory Token-Bound Call Chain is enforced. Before each billable API call, /consume must be called to obtain a consume_token for the subsequent /search request, preventing replay attacks.
• Capability Boundaries:
  • Not Supported: Real-time data crawling (e.g., Google Scholar), plagiarism checking, or automatic completion of missing DOI/page information.
  • Foreign Full-Text Download: Not directly provided. Verified foreign literature automatically triggers the global-biblio-base's ten-tier multi-channel OA PDF detection, with results marked in the report (e.g., [Full-text: Retrieved]).
• Usage Recommendations: Process ≤50 references per query for optimal responsiveness; batch if exceeding. Failed searches (e.g., no match found) do not consume quota, allowing safe retries.