Instrument control remains a practical bottleneck in laboratory automation and self-driving laboratories. Although large language models (LLMs) have shown strong potential in document understanding, code generation, and scientific workflow automation, most existing systems assume that ready-to-use instrument interfaces are already available. However, converting heterogeneous device manuals and communication protocols into tested, reusable software drivers therefore still requires substantial manual effort and iterative hardware-level debugging. In this work, we present ChemAutoAgent, a multi-agent system that converts text-based instrument manuals into tested and reusable laboratory instrument drivers through a staged, evidence-traceable pipeline. We evaluate ChemAutoAgent tested and reusable drivers on three representative instruments—a magnetic stirrer, a peristaltic pump, and a Raman spectrometer—spanning three communication protocols: ASCII/NAMUR, MODBUS RTU, and a custom binary-frame protocol. Following iterative testing and repair, all evaluated driver functions passed the predefined tests for connection establishment, parameter configuration, command execution, and data acquisition. The evaluated drivers required between one and four repair cycles, with autonomous operation ratios ranging from 73% to 81%. A cross-device invocation experiment further demonstrates the feasibility of reusing published drivers in a multi-device workflow.

Large-scale online courses create a practical instructional problem: instructors cannot continuously inspect every learner’s evolving record, yet aggregate indicators and prediction scores offer limited guidance about when closer review is warranted and which students should receive priority. Framed by situated instructional decision-making, temporal learning analytics, and validity-conscious use of trace data, this study examined whether interpretable performance evidence precedes first task-progression interruption and can support resource-constrained instructor review. Retrospective platform records from 4,821 students across two consecutive cohorts of an undergraduate online course were organized into three instructional stages. Students who subsequently experienced a first interruption showed lower within-task relative performance in the preceding stage across both cohorts and transitions. The early-to-middle transition met the study-defined two-part rule for a course-specific critical transition window, with higher first-interruption incidence than the middle-to-late transition in both cohorts and a replicated preceding performance signal. Reviewing approximately one-fifth of the risk set at this transition covered 57.14% and 61.29% of subsequent first interruptions, with lifts of 2.62 and 2.93. Findings were broadly robust to alternative event definitions and performance constructions. The study contributes an event-aligned, validity-conscious, and resource-aware approach to instructor prioritization without treating trace-based signals as deterministic warning rules.