Fact-Check with Multiple AI Models

research

Fact-Check with Multiple AI Models is a Claude Code skill that checks claims, numbers, citations and AI-generated output for hallucinations by cross-checking independent model families: Claude, Gemini and GPT. Trigger it with check-skill-solo or phrases like verify, fact-check or double-check this. It splits the text into atomic claims, searches primary sources through Claude, and in parallel asks the Gemini CLI and the Codex CLI. Every retained source is verified mechanically: the URL must respond and the exact quote must appear on the page. Claims about CLI flags and APIs are accepted only from help output or official docs. The report tags each item as confirmed, consensus-only, disputed, hallucination or unverifiable, with a confidence map. Without Gemini and Codex it runs in single-family mode and says so loudly, because model agreement is then not independent. API keys are optional.