Hivemind Optimizes Agent Skills with the Introduction of SkillOpt

Hivemind, an open-source coding agent skill layer, has introduced SkillOpt, which improves agents' execution ability in real time. While the existing approach simply stored and accumulated agent activity, the new approach evaluates and optimizes the validity of stored skills.

SkillOpt is a text-space optimizer technology developed in collaboration with Microsoft, and works as follows:

• It scores sessions in which a skill was actually used.
• It keeps edits that help improve performance and removes those that do not.
• Since it does not fine-tune the model itself, it adds no inference cost.

Experimental results showed a +19.1 point accuracy improvement in the Claude Code environment, winning or tying in most of 52 test configurations. This feature is provided under the Apache 2.0 license and can be hosted directly within a company's own infrastructure, allowing it to be used while maintaining data security.