verify-claims

Fact-checking a draft for verifiable claims works by extracting citations, numerical facts, and named entities, then spawning a fresh claim-verifier context for independent verification. You receive a structured report classifying each claim as supported, contradicted, or unverifiable.

Chain-of-Verification (CoVe) for academic writing operates by identifying factual claims in a text block and spawning a dedicated verifier with a fresh context to check them independently. It protects draft secrecy while delivering a structured verification report.

Yes, you can supply optional source pointers alongside your draft to guide the fact-checking process. The verifier uses these sources to independently verify extracted claims and report whether they are supported, contradicated, or unverifiable.

Yes, the claim-verifier requires access to the claim-verifier definition and the .claude/agents directory to function. This setup allows it to fork a fresh context for independently verifying extracted claims from your draft.

Post-hoc verification supports configurable failure behavior to handle unverifiable claims. You can define how the system responds when the independent verifier cannot confirm or contradict extracted claims within the structured verification report.