News Fact Check

Problem Addressed

When evaluating a news item, screenshot, or forwarded message, engineers often need a quick trust assessment, but manual checks can miss weak sourcing, recycled stories, or conflicting data. news-fact-check turns scattered suspicion into a reviewable verification workflow. Its focus is not to force a subjective verdict, but to break the claim into inspectable evidence.

How It Works

The skill follows a five-step process:
- Claim decomposition: extract 5W1H, source metadata, core assertions, key figures, and cited authorities, then mark what needs verification.
- Source assessment: rate reliability by tier, from official institutions and major newsrooms to portals and anonymous accounts, while checking red flags such as sensational titles, vague sourcing, and pressure to share.
- Multi-source validation: compare Chinese and English coverage, and use domain-specific authorities such as USGS, WHO, exchange filings, or official statements for earthquakes, health, finance, legal, and international events.
- Contradiction analysis: separate normal differences in measurement, rounding, or timezone from conflicts in core facts, and log source-level discrepancies.
- Report output: return a credibility rating and supporting rationale, including credible, partially credible, doubtful, false, or inconclusive.

Boundaries

It works best for news claims with clear assertions and searchable sources. It is not a substitute for legal review, forensic image analysis, or real-time public opinion monitoring. When evidence is insufficient, the workflow should mark the result as inconclusive rather than forcing a conclusion.