About this project
In one sentence, what are you building?
Crashtest is a tool that takes an AI agent's system prompt, generates and runs a suite of adversarial tests grounded in real documented failures, and reports how - and how badly - the agent broke.
What problem are you solving?
Teams ship AI agents with no real tests, because writing adversarial tests by hand is slow and requires already knowing how agents fail - so agents get judged on whether their responses sound right instead of whether they are right, and break in production in ways nobody anticipated.