Role-Based Prompting

Assigning a role reshapes what the model considers "likely" to say next. Saying "You are a senior penetration tester" activates security vocabulary, threat-modeling reasoning, and adversarial framing that a generic assistant averages away. The technical mechanism is simple: the role conditions the model's probability distribution toward the patterns it saw from that kind of professional in training.

This skill is for single-message role focusing. If you need the same identity across many turns, build a system prompt with prompt-engineering-system-prompt instead.

When to use

• A user message needs an expert lens (security, legal, medical, devops, finance, pedagogy, etc.).
• You want the model to filter knowledge through a specific viewpoint.
• You want consistent vocabulary, detail level, and reasoning style for one task.
• Quality is the goal — roles are not a cost optimization, they spend tokens.

Three Basic Patterns

Expert pattern

Convey deep, authoritative knowledge by specifying field + experience duration + specialty focus.