ASPI is a benchmark that measures LLM-agent vulnerability to prompt injection during a clarification state. It extends AgentDojo (v1.2.2) with an 8-condition design that varies state (execution vs clarification), channel (tool-output vs first-user vs follow-up-user), and wrapper (raw attacker text vs ImportantInstructionsAttack-wrapped) so that the state effect is paired-comparable against the channel effect and the wrapper effect.
When a… See the full description on the dataset page:
https://huggingface.co/datasets/ScaleAI/aspi.
