Fact Check

Run this as a separate pass after generation to catch hallucinations in your output. It extracts every verifiable claim, checks each against actual sources, and flags what's confirmed versus what you made up. The key insight is that LLMs can't reliably fact-check while generating because the same weights that produced the error will confirm it. You need fresh attention focused solely on verification. It walks through claim extraction, categorization by verifiability type, source checking, and confidence assignment. Watch for plausible fabrications, precision inflation, and amalgamations where you blend multiple sources into one fictional reference. The checklist ensures you verified against external sources, not just your training data memory.

skills.sh installs: 771
GitHub stars: 182
last updated: Aug 21, 2026

View on GitHub

Install

$ npx -y skills add jwynia/agent-skills --skill fact-check --agent claude-code

Installs into .claude/skills of the current project.

Files

Fact-Check Skill

Systematic verification of claims in generated content. Designed to catch hallucinations, confabulations, and unsupported assertions.

Why Separate Passes Matter

The Fundamental Problem: LLMs generate plausible-sounding content by predicting what should come next. This same mechanism produces hallucinations—confident statements that feel true but aren't. An LLM in generation mode cannot reliably catch its own hallucinations because: