Loading workspace…
Loading workspace…
Read first — you never hand-write code here.
Your agent writes solution.* on its own computer. The workspace file panel is a read-only, revisioned view: every agent turn flushes the newest revision there. You direct, review, and test — the agent types.
Checking for attempts to resume…
Sample preview. Each attempt gets its own brief and example values. Your exact requirements appear when you start an attempt.
Audit a room calendar with a capacity limit and flag every meeting past capacity, in sweep order. With capacity 1 and meetings [9, 11], [10, 12]: the second starts while the first still runs, so index 1 is flagged.
The billing team shares rooms with capacity 1, and too many meetings got double-booked. Implement overbookedslots(starts, ends, capacity): given parallel lists of meeting start and end times, return the indices of the meetings that push occupancy over capacity. Sweep events in time order with ends before starts at equal times, and flag a meeting when the occupancy at its start, itself included, exceeds 1. Sample: overbookedslots([2,3,5], [7,8,8], 1) -> [1, 2].
def overbooked_slots(starts, ends, capacity):
# your code here
pass
Each attempt gets its own brief and example values. Exact stage 2–3 requirements appear in the workspace as you unlock them.
Checking your paired computers…
Models come from your installed agent. Local provider usage and billing apply.
Solving with: no computer selected · no provider selected · agent default model. Change these above before starting.
Open your workspace with its own question variant, agent chat, and solution file. Write your first prompt there when you are ready.
Starting sets up the workspace. Your agent begins solving only after you send a prompt on the attempt page.
Loading your attempts…
History belongs to this browser session. Reconnecting a computer does not delete it. Archive retains the attempt and its trace. Delete hides it in this browser only.