Adaptive skill libraries lift a fixed computer-use stack 5.7 to 18.6 points, conditionally.

An online skill-evolution framework turns interaction trajectories and evaluator feedback into a persistent versioned library, with each iteration executing against a frozen snapshot so evidence-guided updates only reach later iterations and no model parameters change (arXiv 2609.04869). Against a configuration-matched empty-library control across four OSWorld domains with identical action-generation and grounding stacks, the evolving library won all four post-warm-up runs. Provenance analysis in GIMP found skills retrieved across task-of-origin boundaries and revision churn where repeated accepted edits failed to recover the originating task, so the gains are not monotone. Code at Skill-Evo4GUI.