Bootstrap agent skills AND their verification anchors from docs and repos, not just from task success (OpenSkill)

arXiv:2606.06741·medium signal

OpenSkill (arXiv:2606.06741, submitted June 6, 2026) studies open-world self-evolution where an agent must build both new skills and its own verification signals from scratch, acquiring verification anchors from documentation, repositories, and the web and synthesizing them into transferable skills. This addresses the common failure mode where self-evolving agents reinforce skills they can't actually check. Actionable: when auto-generating a new skill, also extract a checkable verification anchor (a doc invariant, a test, an API contract) so the skill can self-validate before reuse.