Skill Eval Harness 0.5.0 released:

Lots of new features: but the highlight is that it now has proper Codex support. Mistral and Gemini support are on the roadmap.

You run evals against the combination of harnesses/agents and models that you care about then use the results to tweak your skill's trigger rate in all of them. Then you repeat this every time a new…

A healthy long-term approach to evals combines: hill-climbing, valley dodging and floor-raising.

That thing where you're about to make a joke about being the only user of a tool you made and then you discover it somehow has 43 stars: Time to start doing proper releases.