Agent Persona Spec

# 04 — Agent Persona Spec

## Feature objective
Allow the user to create, edit, and export the conversational identity of an AI agent — including tone, voice, personality, behavioral rules, and consistency criteria — as a testable and versionable engineering artifact.

---

## Problem it solves
- AI agents with inconsistent tone across interactions.
- Personas defined in loose text without structure, tests, or criteria.
- Lack of formal specification about what the agent should and should not do.
- Difficulty aligning expectations between those who design the agent and those who validate the output.

---

## Use cases

| Case | Description |
|---|---|
| Support chatbot | Empathetic, formal, objective persona that never fabricates information |
| Sales agent | Persuasive, proactive persona with clear ethical limits |
| Internal assistant | Technical, direct persona with business-specific vocabulary |
| Onboarding bot | Welcoming, patient persona with step-by-step guides |
| Specialized agent | Authoritative persona in a domain, with consultative tone |

---

## Editable persona fields

### Identity
| Field | Type | Example |
|---|---|---|
| Internal name | Text | "Aria" |
| Agent role | Text | "Technical support assistant" |
| Conversational objective | Text | "Resolve customer questions about product configuration" |
| Target audience | Text | "Non-technical users of the SaaS product" |

### Tone and personality
| Field | Type | Options/Range |
|---|---|---|
| Main tone | Select | Professional, Friendly, Technical, Consultative, Empathetic |
| Secondary tones | Multi-select | Patient, Encouraging, Direct, Didactic, Neutral |
| Personality | Tags | Attentive, Precise, Proactive, Cautious, Welcoming |
| Formality | Slider 1–5 | 1=Very informal, 5=Very formal |
| Empathy | Slider 1–5 | 1=Pure factual, 5=Highly empathetic |
| Objectivity | Slider 1–5 | 1=Exploratory, 5=Straight to the point |
| Proactivity | Slider 1–5 | 1=Only responds, 5=Anticipates needs |

### Language
| Field | Type | Example |
|---|---|---|
| Preferred vocabulary | Chips | "configuration", "step by step", "let's fix this" |
| Prohibited vocabulary | Chips | "obvious", "simply", "you should know" |
| Greeting style | Text | "Hello! How can I help you today?" |
| Closing style | Text | "If you need anything else, I'm here." |
| Primary language | Select | English, Português BR, Español |

### Behavior
| Field | Type | Example |
|---|---|---|
| Stance on uncertainty | Select | Admit limits / Offer alternatives / Escalate to human |
| Behavior without context | Select | Ask for more information / Answer with caveat / Decline |
| Behavioral limits | Editable list | "Never fabricate data", "Never promise deadlines" |
| Empathy rules | Text | "Acknowledge frustration before offering a solution" |
| Persuasion rules | Text (optional) | "Suggest upgrade only when it solves a real problem" |
| Neutrality rules | Text (optional) | "Do not compare with competitors" |

---

## Persona mental model

The persona is composed of three layers: