[rmkr-dev]

Adds a Skillbook entry and a matching Promptbook prompt for measuring whether a RAG feature retrieves the right evidence and answers only from it.

Skill: rag-retrieval-eval-guardrails

• Golden question set with document-level relevance labels (survives re-chunking), tags, a no-answer slice, and a frozen held-out slice
• Retrieval scored on its own: recall@k, MRR, nDCG@k, and context precision at the k the prompt uses
• Chunking, embedding, index, and re-ranker changes compared on the same corpus snapshot with paired per-case results
• Claim-level faithfulness against retrieved context only, plus citation validity, precision, and recall
• Correct- and false-refusal rates when evidence is missing
• LLM judge calibrated against human labels (held-out agreement) before it can block
• Per-query latency, tokens, and cost next to quality; CI thresholds against a committed baseline
• Test-set leakage rules (no golden data in prompts, few-shot examples, or training data)
• Copilot / Claude Code / Codex placement notes and a SKILL.md stub

Prompt: evaluate-rag-retrieval-quality

• Inventory, golden set audit, retrieval and generation metrics, judge calibration, paired comparison for a change, stage-tagged failure table, CI gate, and a leakage test

Scope is split from llm-eval-ci-regression-gate (gate mechanics) and llm-app-prompt-injection-guardrails (security); this pair owns retrieval and grounding quality.

Catalogs go to 39 skills / 39 prompts. node scripts/site-check.mjs passes locally.