Observed arrival · 2026-09-06
Valcore, a workbench for testing agent judges
An open-source local-first tool for authoring evaluators, assembling datasets, and measuring agent behavior before release.
Field notes
The page organizes evaluation around versioned judges, labeled examples, and repeatable validation runs rather than a single prompt-quality check. Its examples include refund requests, missing context, and unsafe requests, with labels such as “Policy + next step,” “Clarifying question,” and “Safe refusal.” A displayed run reports 94.2% agreement across 128 cases and passes a 90% threshold. Installation is offered through Homebrew, uv, or pip, with terminal and CI workflows also described.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue