Replication package for:From Steps to Incidents: How Point-Level Evaluation Distorts Attack Detector Assessment for LLM Agents
收藏官方服务:
资源简介:
This package contains the protocol implementation, the per-step scores for all reported runs, the evaluation scripts, and the per-case localization-audit verdicts. Every number in the paper can be regenerated from the files here.
提供机构:
Zenodo创建时间:
2026-11-02



