dsh-task-estimator

Running a multi-turn Agent session in the DeepSeek Harness web GUI, the hardest part is rarely the task itself; it is not knowing how much longer it will take, how many tokens it has burned, or what the final bill will look like. dsh-task-estimator surfaces all three in the session title bar: hover one button and you get a trend-extrapolated remaining-time and finish estimate, a per-model and time-slot (peak/off-peak) cost breakdown in about CNY, and a real-time duration prediction for the round currently in flight.

The standout feature is the optional AI estimate: a lightweight completion call to the session's current model produces a PERT three-point projection (optimistic, most-likely, pessimistic minutes and expected value), remaining round and token forecasts, a task-complexity grade from XS to XL, and a reliability alert whenever the expected duration exceeds that model's p80 threshold. The price table is editable, refresh intervals accept fractional minutes, the panel appears on hover and pins on click, and touch devices degrade gracefully, all without interrupting your workflow.

If you regularly run long DSH sessions, care about per-model billing, or want a quick gut-check on whether a task is within the stable range of the current model, this plugin gives you a number before you start and a live pace readout the whole way through.