---
name: nancy-evals
description: Use to evaluate Nancy/OpenClaw task quality, catch false completion, missing verification, unsafe external actions, wrong Telegram visibility, routing mistakes, or workflow drift.
---

# Nancy Evals

Run evals when Sam asks to improve Nancy/OpenClaw, after large tasks, before saying system work is done, or when trust/reliability is the issue.

Quick command: `skills/openclaw-operator-pack/scripts/nancy-evals.sh`

Checklist lives at `skills/openclaw-operator-pack/references/evals.md`.

Fail closed: if proof is missing, say exactly what proof is missing and what would verify it.