PrimerAgent DB: Toward a Benchmark Dataset for Agent-Orchestrated Primer Design and Multiplex Assays

We propose \textbf{PrimerAgentDB}, a new, openly shareable dataset that transforms primer/probe design from an ad-hoc exercise into a structured, agent-driven resource. The dataset captures primer/probe triplets and multiplex panels across organisms with rich metadata (organism, genome version, thermodynamic and specificity labels, dimer risks, multiplex compatibility), and it is \emph{living}: entries are automatically re-scored as genome references evolve. Unlike existing tools and static databases, PrimerAgentDB couples an \emph{agentic} workflow (retrieval $\rightarrow$ candidate generation $\rightarrow$ specificity/QC $\rightarrow$ multiplex optimization $\rightarrow$ versioned database updates) with public releases suitable for benchmarking AI models on classification, ranking, generation, and planning tasks in molecular assay design.