Search for a command to run...
OccuBench: Evaluating AI Agents on Real-World Professional Tasks via Language World Models