Alibaba Cloud's community blog published a walkthrough on 23 October for QwenWork, describing how to set up a scheduled agent. It is the first of five posted recipes for the same product.
The mechanism as described: the user writes the job and the cadence in one sentence, for example reading a "Weekly Notes" folder every Monday at 8am and producing a digest of what moved, what is blocked and the top three risks. QwenWork converts that into a scheduled task and displays the schedule, the source folder and the output format it inferred. A test run executes immediately against the real files, so the user sees the actual first digest rather than waiting for Monday. The output shape is then tuned by further sentences, such as bullets only and risks first. Digests can be saved as dated markdown files in the same folder, which makes the series accumulate. The task list shows schedule and status, plus a control to pause the job.
Our takeWhat is documented here is a scheduler plus a folder reader, and the post is honest about that: the agent part is still a sentence, and nothing in it says how a digest is checked against the notes it came from. Recurring work fails by drift rather than by setup error, so a "test it now" run catches the wrong folder but not the ninth Monday that quietly drops the blocked items. The dated archive is the right primitive. What is missing is the comparison: a diff of this week's digest against last week's would turn the archive into a check.