Defense-as-Skill: Evolving Runtime Guard Skill for Skill-Augmented Agents
尚未评估计划受阻指标结果claim-table8-judge-audit
Human audit of the LLM judges confirms high reliability: the data construction judge achieves 94% balanced audit agreement (88% success precision, 100% failure confirmation) and the final evaluation judge achieves 98% balanced agreement (96% success precision, 100% failure confirmation).
来源:source-paper:Table 8, page 19
报告指标与观测值
success_precision
m-t8-construction-prec
论文报告 88 percentage_points
实际观测 — percentage_points
failure_confirmation
m-t8-construction-conf
论文报告 100 percentage_points
实际观测 — percentage_points
balanced_audit_agreement
m-t8-construction-agr
论文报告 94 percentage_points
实际观测 — percentage_points
success_precision
m-t8-evaluation-prec
论文报告 96 percentage_points
实际观测 — percentage_points
failure_confirmation
m-t8-evaluation-conf
论文报告 100 percentage_points
实际观测 — percentage_points
balanced_audit_agreement
m-t8-evaluation-agr
论文报告 98 percentage_points
实际观测 — percentage_points
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。