train-persona

Train LoRA adapters for persona agents to embody defined characters with consistent voice, reasoning style, and grounded knowledge. Complements /create-persona which defines WHO the persona is; this skill trains HOW they respond.

Prompt Iteration Rule (NON-NEGOTIABLE)

Persona voice prompts and reasoning trace templates MUST be validated through /prompt-lab before being baked into training data. NEVER hand-craft persona system prompts in Python strings.

• Before training: /prompt-lab eval the persona voice prompt against ground truth examples
• Comparing voice variants: /prompt-lab compare across models
• Only after prompt-lab validation → proceed to LoRA training

Architecture