pi

To compare Pi and Claude Code outputs, you route build requests through Pi using a single command that automatically injects full vault context. This launches Pi in a new terminal window for side-by-side comparison testing without manually rebuilding the task context.

You route a specific Dex skill through Pi by requesting skill-aware execution, which sends the exact same instructions Claude Code would receive. This ensures the comparison test is isolated to the model's reasoning rather than differences in task context.

Yes, you can enable Pi mode to consistently route all future build requests through Pi for the duration of your session. This prevents the need to manually specify Pi for every individual task launch.

If Pi or the required skill files are missing during a task launch, the environment-aware terminal launch process triggers error handling. This prevents silent failures and alerts you to the missing dependencies needed for context injection.

Yes, context injection works the same way for ad hoc requests, gathering the necessary vault context before launching Pi in a new terminal window. This allows you to test custom prompts side-by-side with Claude Code results.