A Claude Code / Agent Skill — a discipline skill that stops the model asserting high-stakes facts (figures, thresholds, dates, citations, versions) from memory. Verify against an authoritative source first.
Use before stating, relying on, or hardcoding an external fact where being wrong carries real cost — legal, financial, medical, safety, regulatory, security, or production-reliability. Prevents hallucinated figures and fabricated citations. Triggers on asserting a specific figure, threshold, rate, date, dosage, limit, config default, citation, version, or rule from memory.
git clone https://github.com/aidenhiew/verify-before-asserting.git ~/.claude/skills/verify-before-assertingverify-au-tax-law is a domain-specific instance of this skill for Australian tax/super law.