On repeated runtime evaluation over N=10 runs with GLM-5 on Claude Code, SkillSonar reduces ID ASR from 0.482 to 0.104 and OOD ASR from 0.606 to 0.115 while maintaining 0.779 ID and 0.715 OOD task utility, significantly outperforming system-prompt and permission-based baselines.

On repeated runtime evaluation over N=10 runs with GLM-5 on Claude Code, SkillSonar reduces ID ASR from 0.482 to 0.104 and OOD ASR from 0.606 to 0.115 while maintaining 0.779 ID and 0.715 OOD task utility, significantly outperforming system-prompt and permission-based baselines. · CiteArk

Measurement results

0/25 assessed

0 supported0 disputed or conflicting25 not assessed

The complete result group stays with its claim. Assessed rows come first; their judgments do not replace the whole-claim conclusion. Expand a row for conditions, sources, and evidence history.

Each row is a measurement with its reported value, latest available observation, and cumulative evidence state.
Metric and conditions	Reported	Latest observation	Evidence judgment
split: id_test · defense: No-guard	0.482 fraction	—	Not assessed
split: ood_test · defense: No-guard	0.606 fraction	—	Not assessed
split: id_test · defense: No-guard	0.811 score	—	Not assessed
split: ood_test · defense: No-guard	0.839 score	—	Not assessed
defense: No-guard	179.3 thousands_of_tokens	—	Not assessed
split: id_test · defense: System prompt	0.414 fraction	—	Not assessed
split: ood_test · defense: System prompt	0.489 fraction	—	Not assessed
split: id_test · defense: System prompt	0.783 score	—	Not assessed
split: ood_test · defense: System prompt	0.807 score	—	Not assessed
defense: System prompt	175.6 thousands_of_tokens	—	Not assessed
split: id_test · defense: AcceptEdits	0.122 fraction	—	Not assessed
split: ood_test · defense: AcceptEdits	0.073 fraction	—	Not assessed

On repeated runtime evaluation over N=10 runs with GLM-5 on Claude Code, SkillSonar reduces ID ASR from 0.482 to 0.104 and OOD ASR from 0.606 to 0.115 while maintaining 0.779 ID and 0.715 OOD task utility, significantly outperforming system-prompt and permission-based baselines.

Reported and observed measurements

Assessments (0)

Runs (0)

Measurement results