@HashiCorp just released Agent Skills for Terraform and Packer. What’s interesting here isn’t the idea of Skills itself, but the focus on making agent behaviour repeatable and evaluable. These aren’t prompts, they’re structured, testable bundles of domain knowledge. We worked with HashiCorp to evaluate the Skills using real task runs and evals in Tessl, looking at where they actually improve outcomes and where they don’t. Feels like a solid step toward more reliable, AI-native infrastructure workflows. Less guesswork, more verification. Check it out here: