Find LLM output reaching sensitive sinks (OWASP LLM05)

Find freeform LLM output flowing into SQL text, process commands, script evaluation, or PrintWriter output (including servlet responses). Covers text extraction in LangChain4j, Spring AI, OpenAI, Anthropic, and Azure AI Inference. Findings require review of parameterization, validation, and context-specific encoding; PrintWriter output is not necessarily HTML. Tracks flows within the enclosing method through transformations without assuming that arbitrary helpers sanitize output. Does not report prepared-statement bind values as SQL text, or determine whether an upstream sanitizer is appropriate for the destination. No automatic fix is applied.