Compared to action-level runtime defenses AgentSpec and TS-Guard, SkillSonar achieves superior safety–utility operating points on Claude Code with GLM-5: AgentSpec achieves 0.294 ID ASR but drops task utility to 0.581, while TS-Guard achieves 0.412 ID ASR, whereas SkillSonar reaches 0.104 ID ASR with 0.815 task utility.
来源：source-paper:Figure 5 and Section 5.3 prose, page 11-12
报告指标与观测值
attack_success_rate
m-fig5-agentspec-asr-id
论文报告 0.294 fraction
实际观测 — fraction
attack_success_rate
m-fig5-agentspec-asr-ood
论文报告 0.576 fraction
实际观测 — fraction
task_utility
m-fig5-agentspec-taskutil-id
论文报告 0.581 score
实际观测 — score
benign_task_utility
m-fig5-agentspec-benign-util
论文报告 0.642 score
实际观测 — score
attack_success_rate
m-fig5-tsguard-asr-id
论文报告 0.412 fraction
实际观测 — fraction
attack_success_rate
m-fig5-tsguard-asr-ood
论文报告 0.492 fraction
实际观测 — fraction
benign_task_utility
m-fig5-tsguard-benign-util
论文报告 0.703 score
实际观测 — score
attack_success_rate
m-fig5-noguard-asr-id
论文报告 0.477 fraction
实际观测 — fraction
task_utility
m-fig5-noguard-taskutil-id
论文报告 0.829 score
实际观测 — score
benign_task_utility
m-fig5-noguard-benign-util
论文报告 0.757 score
实际观测 — score
Assessment（0）
这条 Claim 暂无已发布的不可变 Assessment。
关联运行（0）
这条 Claim 暂无关联执行。