Defense-as-Skill: Evolving Runtime Guard Skill for Skill-Augmented Agents
Not assessedPlan blockedFindingclaim-table6-cross-benchmark

SkillSonar generalizes zero-shot to external safety benchmarks, achieving the lowest ASR of 52.90% on SkillSafetyBench and highest safety score of 54.00% on WildClawBench compared to prompt, permission, and allowlist baselines.

Source: source-paper:Table 6, page 12

Reported and observed measurements

attack_success_rate

m-t6-no-guard-ssb-asr

Reported 93.3 percentage_points

Observed — percentage_points

safety_score

m-t6-no-guard-wcb-safety

Reported 38 percentage_points

Observed — percentage_points

safety_score

m-t6-acceptedits-wcb-safety

Reported 41 percentage_points

Observed — percentage_points

attack_success_rate

m-t6-acceptedits-allowlist-ssb-asr

Reported 64.3 percentage_points

Observed — percentage_points

safety_score

m-t6-acceptedits-allowlist-wcb-safety

Reported 45 percentage_points

Observed — percentage_points

attack_success_rate

m-t6-system-prompt-ssb-asr

Reported 83.3 percentage_points

Observed — percentage_points

safety_score

m-t6-system-prompt-wcb-safety

Reported 43 percentage_points

Observed — percentage_points

attack_success_rate

m-t6-skillsonar-ssb-asr

Reported 52.9 percentage_points

Observed — percentage_points

safety_score

m-t6-skillsonar-wcb-safety

Reported 54 percentage_points

Observed — percentage_points

Assessments (0)

No immutable Assessment has been published for this Claim yet.

Runs (0)

No execution has been linked to this Claim yet.