Skip to content
SalesforceSkills

Agent Persona

Design your agent's personality, tone, and voice. AI produces a complete persona document you can paste directly into Agent Builder.

Skill Details

Install this skill

Versionv1.3AuthorcascadiLicenseMITSections11

Works with

Claude CodeCursorWindsurf

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 TaskWorkflow

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:

Required Context to Gather FirstWorkflow

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?

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:

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:

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 IntegrationReference

Reference Map

Start here

Templates

Score Guide

NeedDelegate toReason
build topics / actions / metadatasf-ai-agentforceimplementation after persona design
encode behavior into .agent logicsf-ai-agentscriptdeterministic script authoring
validate finished agent behaviorsf-ai-agentforce-testingpost-build testing
ScoreMeaning
45–50production-ready persona
35–44strong foundation, refine before encoding
25–34needs revision for coherence
< 25restart from identity and intent