Let’s start from the end. I’ll tell you how to run it e2e if you like, and then we’ll dig into the code.
If you want to follow along at home, go ahead and echo "ANTHROPIC_API_KEY=ant-sk-..." > .env, and run:
uv run agent dafnybench inspect
This will run 10 samples in Haiku, which is vanishingly small in token cost. You can monitor usage on console.anthropic.com.
Inspect has a TUI, but instead we’re going to open up a new terminal and type uv run inspect view for their webapp logviewer. More likely, I’m going to type uv run inspect view, and you’ll just follow along via screenshot.
Figure 1:Inspect’s log viewer webapp.