Purpose: “LLM-as-judge” prompt templates used by the evaluation harness to score skill outputs (rationale quality, groundedness, extraction accuracy) — distinct from the production prompts in prompts/system/.
What belongs here: Judge-prompt templates with clear scoring rubrics/output schemas.
What must not be stored here: Real candidate data in examples; actual evaluation run results (→ evals/reports/).
Owner: [TENANT_CONFIGURATION_REQUIRED — AI Governance Lead]
Contents: evaluation-prompts.md
Status: Populated.