SkillForge treats skill-library management as an active forging process: a fitness-driven lifecycle moves every skill through trial, active, stable and retired states while the policy trains, retiring what the improved model has outgrown and mutating what has room to improve — up to 7.8% relative gain over the strongest memory-augmented RL baseline, with a compact library and the SkillFurnace dataset of 5,852 annotated records.
