LLMs (especially smaller models) sometimes pass multiple indicator names as a single comma-separated string instead of making separate tool calls. Split and process each individually at the tool boundary.