Persona Fidelity Audit

Scheduled audit of agent persona fidelity. The mechanical scorer (modules/conversations/persona_scorer.py) stamps every assistant message with a persona_score at insert time. This job harvests that data, escalates outliers to the LLM-as-judge sparingly, and writes an evidence-based report. Pairs with the persona-fidelity-audit skill which holds the run-time prompt.

Details

Scheduled audit of agent persona fidelity. The mechanical scorer (modules/conversations/persona_scorer.py) stamps every assistant message with a persona_score at insert time. This job harvests that data, escalates outliers to the LLM-as-judge sparingly, and writes an evidence-based report. Pairs with the persona-fidelity-audit skill which holds the run-time prompt.