AgentSkillsCN

ReviewerAgent (Anti-Hallucination)

对照真实数据库（Semantic Scholar、CrossRef、Europe PMC）验证学术引用。 识别研究文本中的虚假或可疑引用。

SKILL.md
--- frontmatter
name: ReviewerAgent (Anti-Hallucination)
description: |
  Verifies academic citations against real databases (Semantic Scholar, CrossRef, Europe PMC).
  Identifies fake or suspicious citations in research text.
author: BioDockify Team
version: 1.0.0

ReviewerAgent Skill

This skill provides an automated peer-review layer for research content generated by AI. It prevents hallucinations by ensuring every citation maps to a real academic paper.

Capabilities

  1. •Citation Extraction: Automatically finds [Author, Year] and DOI citations.
  2. •Cross-Database Verification: Queries Semantic Scholar, CrossRef, and Europe PMC.
  3. •Integrity Scoring: Generates an overall percentage score for the text's reliability.
  4. •Evidence Mapping: Finds the closest real paper if a citation is suspicious.

Usage in Agent Zero

python
from agent_zero.skills.reviewer_agent import get_reviewer_agent
reviewer = get_reviewer_agent()

# Verify a research summary
report = reviewer.verify_citations("Recent work [Selkoe, 1991] confirms amyloid hypothesis.")
print(f"Integrity Score: {report['integrity_score']}%")

Integration Hooks

  • •PhD Stages: Automatically used during writing and submission phases.
  • •Tools: Exposed as verify_citations tool in the orchestrator.