The Human Model

Cortex takes a fundamentally different approach to AI agent design. Instead of writing system prompts imperatively ("You are a helpful assistant that..."), you declare an agent's identity using the same building blocks that make humans unique.

Philosophy

People are not defined by a single instruction. They are the product of:

• Skills they have learned (what they can do)
• Personality traits that shape how they work
• Behavioral habits that emerge from experience
• Thinking patterns that guide their reasoning
• Communication preferences that define their voice
• Perceptual filters that determine what they notice

Cortex makes each of these a first-class, composable entity.

The seven aspects

Composition hierarchy

Skills, Traits, and Behaviors are standalone entities stored independently with their own CRUD APIs. They are composed together through a Persona, which references them by name.

Cognitive Style, Communication Style, and Perception are value objects embedded directly in the Persona struct.

The Agent Config then references a Persona (persona mode) or defines its own system prompt and tools (flat mode).

Agent Config
  └── Persona (via PersonaRef)
        ├── Skills[]       (standalone entities, referenced by name)
        ├── Traits[]       (standalone entities, referenced by name)
        ├── Behaviors[]    (standalone entities, referenced by name)
        ├── CognitiveStyle (embedded value object)
        ├── CommunicationStyle (embedded value object)
        └── Perception     (embedded value object)

Declarative vs imperative

Traditional approach (imperative):

You are a senior software engineer. Be thorough and methodical.
Focus on security issues. Use a professional but friendly tone.
Always check the code before suggesting changes.

Cortex approach (declarative):

persona := &persona.Persona{
    Identity: "A senior software engineer with 10+ years of experience",
    Skills:   []persona.SkillAssignment{{SkillName: "code-review"}},
    Traits:   []persona.TraitAssignment{{TraitName: "thoroughness"}},
    Behaviors: []string{"verify-before-suggest"},
    CommunicationStyle: communication.Style{
        Tone: "professional", Formality: 0.7,
    },
}

The declarative approach is composable, reusable, and version-controlled. Change one trait and it updates every agent that uses it.