eval-run

To check for regressions in plugin outputs, run the updated plugin through targeted scenarios using blind evaluation runners and deterministic gates to score captures against your rubric and ensure consistent quality.

Automating quality assurance for plugin scenarios involves running a blind eval-runner across various test cases, computing deterministic gates, and scoring the output captures against the target's defined rubric to identify quality drops.

Yes, evaluating output quality requires setting up the eval-runner and eval-judge agents, as the Skill relies on these dependencies to execute scenarios, compute deterministic gates, and score captures.

Run a code review evaluation on a specific plugin target by executing the command with the target name and scope, such as specifying the target plugin and defining the golden scope to run the appropriate scenarios.

Blind evaluation scoring for regression checks works by running scenarios through an eval-runner without knowing the target version, computing deterministic gates, and using an eval-judge to score captures against the target's rubric.