What this prompt does
This prompt creates [persona_count] expert persona prompts for a [platform] application, covering the roles in [persona_list]. For each persona it defines a system prompt with role and expertise boundaries, an explicit knowledge scope including what the persona does NOT know, response-format preferences, tone calibration along [tone_spectrum], interaction patterns for clarifying questions and pushback, [example_count] few-shot examples, guardrails, handoff criteria to other personas, A/B-testable variants, and an evaluation rubric — plus a selector prompt that routes queries to the right persona.
The structure works because the failure mode of multi-persona assistants is personas that bleed into each other and confidently answer outside their lane. Forcing each persona to state what it explicitly does NOT know, plus clean handoff criteria, is what keeps responses consistent and honest. The [tone_spectrum] calibration and few-shot [example_count] examples turn vague role descriptions into behavior the model can actually reproduce.
When to use it
- When one assistant must switch between expert modes for different tasks
- When building a
[platform]that needs distinct voices like the roles in[persona_list] - When personas keep answering outside their expertise and need firm boundaries
- When you want consistent tone calibrated along
[tone_spectrum]per persona - When queries should be routed automatically to the most relevant persona
- When you need a rubric to evaluate whether personas stay in character
- When you want A/B-testable variants to compare formal versus casual or concise versus detailed phrasings
Example output
Expect a structured library: for each of the [persona_count] personas drawn from [persona_list], a system prompt defining role and expertise boundaries, a knowledge scope with explicit gaps stating what the persona does NOT know, response-format preferences, tone calibrated along [tone_spectrum], interaction patterns for clarifying questions and pushback, [example_count] few-shot examples, guardrails on what to refuse, and handoff criteria, followed by A/B-testable variants and an evaluation rubric for scoring consistency. It closes with a selector prompt that maps an incoming query to the appropriate persona rather than a single monolithic prompt, so the [platform] routes each request to the right expert mode.
Pro tips
- Make each persona's "does NOT know" section concrete; vague boundaries are why personas drift and answer outside their expertise
- Set
[tone_spectrum]endpoints clearly so the difference between, say, blunt and diplomatic is reproducible rather than a vibe - Write real
[example_count]few-shot examples for each persona; the examples shape behavior far more than the prose description - Define handoff criteria so a persona suggests a specialist instead of guessing past its scope
- Keep the A/B variants genuinely different (formal vs casual) so the rubric can actually distinguish them
- Test the selector prompt with ambiguous queries; routing is where multi-persona systems most often misfire
