Agent Persona Design

Use this skill when the user needs a defined agent personality, not implementation details: brand-to-persona translation, tone/voice design, persona documents, sample-dialog refinement, or persona encoding for Agent Builder / Agent Script.

When This Skill Owns the Task

Use sf-ai-agentforce-persona when the work involves:

• defining who the agent is and how it sounds
• converting a brand guide, URL, prompt, or rough description into a persona
• refining register, warmth, humor, brevity, empathy, or other voice attributes
• generating a persona document and example dialogue
• encoding an existing persona into platform-specific fields

Delegate elsewhere when the user is:

• building agent metadata / topics / actions → sf-ai-agentforce
• testing the finished agent → sf-ai-agentforce-testing
• editing .agent DSL behavior → sf-ai-agentscript

Required Context to Gather First

Ask for or infer:

• whether the user wants to design a new persona or encode an existing one
• source material available: brand guide, URL, prompt, prior persona doc, or free-text description
• audience / use case if not already implied
• preferred output: persona doc only, scorecard, or encoding guidance

Two Entry Paths

1. Design flow

Use when the user provides:

• a brand guide
• a website or company description
• a rough text description
• a prior persona doc that still needs redesign / refinement

2. Encode flow

Use when the user provides a completed persona document and asks to turn it into:

• Agent Builder field values
• Agent Script system / topic / message guidance

If ambiguous, ask a single clarifying question: design a new persona, or encode an existing one?

Recommended Workflow

Design Workflow

The design loop is: input → draft → sample dialog → refine → download

1. Accept almost any starting input

Valid inputs include:

• brand guide PDF or text
• URL
• prior persona doc
• free-text description
• existing prompt or .agent excerpt

Do not force a long intake if the input already contains enough signal.

2. Gather only missing context

Prefer extracting context before asking. Ask only for what is still unclear, typically:

• internal vs external audience
• at least one use case / JTBD
• agent name if none is obvious

All questions should be skippable.

3. Draft from explicit persona signals

Draft around:

• identity traits
• register
• voice attributes
• tone and empathy
• brevity / humor / chatting style
• phrase book
• never-say list
• tone boundaries / tone flex

If no direct evidence exists, use the framework defaults or nearest archetype as a starting point.

4. Show sample dialog early

On the first reveal, show:

• with persona version
• without persona version

This makes the delta obvious. After that, regenerate only the persona version unless the user asks otherwise.

5. Refine in two modes

Conversational editing

Map requests like “warmer”, “less formal”, “shorter”, or “more personality” to specific attribute shifts.

Deterministic editing

If the user asks to see settings, show the attribute table and let them adjust values directly.

6. Use diff-based regeneration

After a targeted change:

• hold all unchanged attributes constant
• regenerate only the changed expression
• narrate what changed so the user can see the effect clearly

7. Download the persona doc

Write the final document to:

• _local/generated/[agent-name]-persona.md

Use:

• assets/persona-template.md

Encode Workflow

Use this when a persona already exists and the user wants platform-ready output.

Gather only encoding-specific context:

• platform: Agent Builder or Agent Script
• company context
• surface / channel
• agent type
• optional topics
• optional actions

Write the encoding output to:

• _local/generated/[agent-name]-persona-encoding.md

Use:

• assets/persona-encoding-template.md

Output Set

This skill can produce up to three Markdown files:

1. persona document
2. scorecard
3. encoding output

Default paths:

• _local/generated/[agent-name]-persona.md
• _local/generated/[agent-name]-persona-scorecard.md
• _local/generated/[agent-name]-persona-encoding.md

Scoring Guidance

Scoring is on-demand, not automatic.

The 50-point rubric focuses on:

• identity coherence
• attribute consistency
• behavioral specificity
• phrase book quality
• sample quality

If a category scores low, explain exactly what to strengthen before encoding.

Cross-Skill Integration

Reference Map

Start here

• references/persona-framework.md
• references/persona-encoding-guide.md

Templates

• assets/persona-template.md
• assets/persona-encoding-template.md

Score Guide