| description | Execute an eval YAML produced by /eval — runs against a real target, scores metrics, writes summary.md + metrics.csv. HARD-STOPS if no target configured (no fake results). |
|---|---|
| argument-hint | [path-to-eval-yaml] [--only id1 id2] [--max-tokens N] [--judge-model NAME] [--dry-run] |
Invoke the pmstack-run-eval skill from the pmstack plugin using the Skill tool, passing these arguments verbatim:
$ARGUMENTS
Follow that skill exactly. Do not improvise a lighter-weight version of it. If the skill is unavailable, say so and point the user to /plugin install pmstack@pmstack — do not attempt the task from general knowledge.