EvoSkill, an open-source toolkit which takes a benchmark and a coding agent (like Claude Code), then evolves that agent into a state-of-the-art specialist on that task automatically.
Just released our newest piece of research that identifies gaps in a coding agent's operation on various capabilities, and then creates the skills needed to fill those gaps. Our preliminary tests show this evolutionary loop leads to substantial improvement on the hardest benchmarks. Also, we see skills as a powerful abstraction that can be leveraged for generalizability---the skills created in this loop also lead to improvement in benchmarks outside of the direct one tested.
Report
No reviews yetBe the first to leave a review for EvoSkill
ROMA