Agenthog add eval — PLANNED

This skill is planned but not yet written. The intended scope:

1. Pick an evaluator template (hallucination, helpfulness, format-correctness, custom LLM-as-judge).
2. Wire the eval to run on a sample of incoming traces (e.g. 10%) or on a dataset on demand.
3. Surface the eval score as a agent.eval event on the trace.
4. Verify the score shows up in the AgentHog Evaluators tab.

Track progress at https://github.com/TheAgentOS/agentos-skills/issues