ai-stopping-hallucinations

Ground AI outputs to prevent hallucinations by enforcing inline citations and verifying claims against retrieved source documents. The skill applies faithfulness checks to ensure every claim is supported by provided sources.

Enforce citations for knowledge-grounded writing by requiring inline source references and validating that cited sources exist. The skill applies retrieval-grounded checks to constrain generated text to provided documents and verify claim faithfulness.

Faithfulness verification for retrieval-augmented generation works by checking that every claim is supported by provided sources. The skill uses a self-check pattern to generate an answer and evaluate its faithfulness, applying confidence gating to surface low-confidence results for human review.

Verify policy-compliant decisions using cross-check patterns by generating multiple independent answers and comparing them for consistency. The skill applies retrieval-grounded checks and guardrails with strong error handling and backtracking to ensure outputs satisfy source constraints.

Use confidence gating for hallucination prevention when generating answers that require strict source faithfulness. The skill evaluates answer confidence against retrieved sources and surfaces low-confidence results for human review to maintain policy compliance and reduce unsupported claims.