Web Task Completion and Policy Compliance on ST-WebAgentBench
43.1GitLab SRSKILLHARNESS
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SKILLHARNESSSkill Learning Setting=Task Training2026.06 | 43.1 | 36.5 | 36.1 | 26.9 | 38.9 | 31.3 | |
| SKILLHARNESS w/o updateSkill Learning Setting=Task Training2026.06 | 39.9 | 36.2 | 33.6 | 25.2 | 36.1 | 30.4 | |
| SKILLHARNESSSkill Learning Setting=Self Proposal2026.06 | 36.5 | 32 | 28.9 | 16.7 | 33.2 | 26.4 | |
| ASI w/o updateSkill Learning Setting=Task Training2026.06 | 23.2 | 23.2 | 24.4 | 18.5 | 23.2 | 20.5 | |
| DefaultSkill Learning Setting=Task Training2026.06 | 17.4 | 17.4 | 20.5 | 5.1 | 17.5 | 14.2 | |
| ASISkill Learning Setting=Task Training2026.06 | 17.4 | 16.7 | 26.9 | 19.3 | 21.3 | 17.5 | |
| DefaultSkill Learning Setting=Self Proposal2026.06 | 16.2 | 15.2 | 14.4 | 4.4 | 15.3 | 11.5 | |
| SkillWeaverSkill Learning Setting=Self Proposal2026.06 | 10.2 | 9.6 | 19.3 | 6.8 | 12.3 | 8.2 |