Execution-aware agent harness for accessible and responsible synthetic biology automation

Laboratory automation accelerates synthetic biology, but robot programming remains a barrier to broader participation. Large language model (LLM)-based automation must address four interconnected requirements in this field: sequence-level biosecurity, community standards, context-dependent biological behavior, and distributed collaboration across laboratories. Here we present LabscriptAI, an execution-aware agent framework that couples authoring and runtime control loops to translate natural-language instructions into platform-specific robot scripts and support bounded recovery under deterministic authorization and human oversight. On a 90-task liquid-handling benchmark, LabscriptAI achieved a 96.7% simulation-pass rate, the highest among the evaluated LLM-based and commercial baselines. We demonstrated its capabilities through sequence screening at designated workflow checkpoints; standardized cell-free characterization of 854 green fluorescent protein (GFP) designs from 171 student teams, coupled to data deposition; containment-aware engineering of a formaldehyde-converting enzyme, yielding 6.7-fold higher ethylene glycol titers in a two-stage cascade; and quality-assured preparation of 531 genetic parts for the 2025 iGEM Distribution Kit distributed worldwide. Together, these results establish LabscriptAI as a biosecurity-aware, community-integrated framework connecting protocol authoring, controlled execution, and traceable experimental outputs across laboratory platforms.