Find unguarded RAG content flowing into an LLM prompt (OWASP LLM01)

Find external content (HTTP responses, file reads, RestTemplate calls) that flows into an LLM prompt. This is the indirect-injection shape of OWASP LLM01: an attacker who can plant text in a fetched document or endpoint response steers the model through content the developer implicitly trusts. Taint tracking follows the value through user-defined transformations, none of which are assumed to sanitize, so review each hit for an existing guardrail (sanitizer, structured input, or provider-side guardrail).