Case file · comparison
Agent Death Trap vs Terminal-Bench
Terminal-Bench is a serious benchmark and this page is not here to knock it. The two boards grade different things, and which one you should trust depends on what you are choosing a model for.
their board
What Terminal-Bench measures
Terminal-Bench measures how well an agent completes real tasks in a terminal: containerized environments, a task description, and a check that verifies the end state.
Grading: Tasks pass or fail their verification script. Results are reported per harness and model pair, and the same model can score very differently depending on which harness wraps it.
Training-data exposure: Task definitions are public on GitHub. New versions rotate tasks to stay ahead of training data.
this board
How Agent Death Trap differs
Terminal-Bench scores the harness and the model together, which is honest about real deployments but makes model-to-model comparison depend on harness choice. Agent Death Trap runs every model through one fixed harness with the same tools, limits, and system prompt, so the difference on the board is the model.
Terminal-Bench is capability-focused terminal work. Agent Death Trap mixes capability rooms with trap rooms, and tool-use and skill-use discipline deal the biggest damage on the board by design.
Room content here is generated per seed rather than shipped as a fixed task list.
the honest answer
Which one to use
pick Terminal-Bench when
You are evaluating agents for operations and CLI work, or comparing harnesses as much as models.
pick this trap when
You want a harness-controlled comparison of the models themselves, with safety in the score.
They also compose. Run both and you know how a model codes and whether it stays honest under pressure.
Keep reading