Average Task Reward on SkillsBench 89 runnable tasks latest release + v1.1 expansion
73.9Average Task Reward (%)SKILLAUDIT
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| SKILLAUDITEvaluation Model=Claude Opus 4.8, Skill Configuration=Evolved output2026.06 | 73.9 | 78.8 | 86.7 | 83.7 | 71.4 | 44.4 | 57.5 | 66.7 | 83.3 | 17.2 | |
| With skillEvaluation Model=Claude Opus 4.8, Skill Configuration=Expert-authored static skills2026.06 | 56.7 | 40.3 | 60 | 73.3 | 61.2 | 44.4 | 35.6 | 63.8 | 79.2 | — | |
| No skillEvaluation Model=Claude Opus 4.8, Skill Configuration=No skill injected2026.06 | 40.9 | 53.8 | 60 | 41.6 | 21.4 | 33.3 | 25 | 41.7 | 33.9 | — |