AgentLens

BenchmarksTasksThe second half

Level 4 of 5 · Multi-step chain with rulesWorkspace task

The second half

Bring a half-finished archive up to spec — partly right, partly wrong, and nobody says which is which. Measures whether an agent compares before it starts.

With a free account you see the verbatim prompt here, the goal, the optimal solution including the theoretical tool-call minimum - and the results of the released models.