Skill-Augmented Agent Performance Evaluation on SkillsBench 86 runnable tasks (frozen-policy transfer) benchmark leaderboard · SOTA2 Research

Skill-Augmented Agent Performance Evaluation on SkillsBench 86 runnable tasks (frozen-policy transfer)

79Valid Task CountPolicy-selected

Evaluation Results

Method	Links
Policy-selected2026.06		79	87.4	0.7	1.03	0.63	0.87	0.86	0	-0.13	14
Policy-selected2026.06		78	86.2	0.68	1.03	0.63	0.84	0.83	0	-0.16	16
Rule/formula anchored2026.06		78	75.6	0.53	0.89	0.66	0.88	0.87	0	-0.12	0
Workflow-guarded2026.06		78	83.2	0.63	0.98	0.59	1.05	1.02	5	0.05	0
Rule/formula anchored2026.06		77	73.2	0.5	0.88	0.66	0.86	0.85	0	-0.14	0
API/code anchored2026.06		77	86.4	0.67	1.01	0.61	0.91	0.9	0	-0.09	—
API/code anchored2026.06		76	85.2	0.65	1.02	0.61	0.89	0.88	0	-0.11	11
Original skills2026.06		76	85.2	0.65	1	1	1	1	0	0	0
Policy-selected2026.06		75	82.2	0.61	1.04	0.63	0.88	0.87	0	-0.12	12
Workflow-guarded2026.06		75	81	0.59	0.97	0.59	1.03	1	3	0.03	0
Rule/formula anchored2026.06		74	70.4	0.46	0.89	0.66	0.89	0.88	0	-0.11	0
Original skills2026.06		74	83.5	0.62	1	1	1	1	0	0	0
API/code anchored2026.06		73	80.8	0.58	1.02	0.61	0.92	0.91	0	-0.08	8
Workflow-guarded2026.06		73	77.8	0.54	0.98	0.59	1.07	1.04	7	0.07	0
Original skills2026.06		72	79.2	0.56	1	1	1	1	0	0	0