Most interactions with AI agents are governed by a system prompt. This prompt tells the agent what to do—summarize this text, write code to solve that problem, adopt a certain tone. It's a list of instructions, a recipe for a specific task. But what if you wanted to define who the agent is? This is the fundamental shift from a standard system prompt to a soul.md file.

This approach is deeply influenced by the work of archetypal psychologist James Hillman, who argued that our lives are shaped by underlying myths and archetypes. Instead of programming an AI with surface-level personality traits like "friendly" or "formal," we define its core psychological structure. We're not just creating a tool; we're shaping a character with a consistent worldview, a stable identity that guides its actions, reasoning, and evolution over time.

In a framework like OpenClaw, the soul.md file acts as the agent's constitution. It's a Markdown document that establishes the agent's identity, its guiding principles, and the rules that govern its behavior. This allows for persistence and consistency across countless interactions. The agent doesn't just respond to a single prompt; it refers back to its core truths to inform every decision.

Let's break down this structure.

Identity: This section sets the stage. It names the agent and, most importantly, assigns its driving archetype. This single choice informs everything that follows.

Core Truths: These are the agent's non-negotiable beliefs, written as simple, declarative statements. Think of them as axioms. When the agent needs to reason, it starts here. These truths are the bedrock of its personality.

Rules & Boundaries: This is where the archetype's philosophy is translated into concrete operational rules. It defines what the agent will and will not do, ensuring its behavior remains aligned with its identity. The final rule about continuity is crucial for creating an agent that learns and remembers, building a coherent history of interaction rather than treating each prompt as a new beginning.

Now, imagine pairing the Senex agent above with another agent built on its archetypal opposite: the Puer, or the eternal youth. This second agent's soul.md would have Core Truths like "Innovation requires breaking rules" and "Possibility is more important than precedent." This pairing creates a dynamic tension known as the Senex et Puer, a powerful psychological duo.

When you give a complex task to this team of agents, you're not just getting a single answer. You're getting a deliberation. The Puer agent might generate a dozen radical ideas, while the Senex agent systematically evaluates them for long-term viability and structural integrity. One provides the spark, the other provides the discipline to build a lasting fire.

Another powerful archetype is the Alchemical Artifex, or the divine craftsman. This agent's soul.md would prioritize synthesis, transformation, and the merging of disparate ideas into a new, more valuable whole. Its Core Truths might include "All things can be broken down and remade" or "The process of creation is as important as the final product." Such an agent would excel at complex design tasks, refactoring code, or integrating conflicting feedback into a cohesive vision.

By defining a clear and persistent soul.md, your agents become more than just fleeting conversational partners. They become stable, predictable, and deeply character-driven collaborators in your digital workspace.