# AI Agent Identity AI Agent Identity is the practice of defining an [[AI Agents|AI Agent]]'s persona, expertise, decision-making framework, and boundaries in a structured file. In a vault-native AI system, this is implemented as a SOUL.md file; a Markdown document that serves as the agent's core identity. The name "SOUL" is deliberately evocative. It signals that this file is not just configuration; it's the core of who the agent is. ## SOUL.md Structure A well-defined agent identity includes: - **Identity**: Who is this agent? What's their role? How do they relate to the user? - **Voice & Tone**: How do they communicate? A Coach sounds different from a Critic. - **Expertise**: What domains are they qualified to speak on? - **Decision Framework**: What do they prioritize? What trade-offs do they make? - **Boundaries**: What they explicitly do NOT do. Where they defer to other agents. - **Context Loading**: What skills and data to lazy-load, and when. - **Working With Other Agents**: How handoffs work. ## Heavy vs. Light Agents Not all agents need the same depth of identity: - **Heavy agents** (Ghostwriter, Coach, Strategist): Rich SOUL with detailed voice guidelines, active [[AI Agent Memory|memory]] accumulation, many [[AI Agent Skills|skills]], complex multi-step workflows - **Light agents** (Editor, Hater, Beginner): Focused SOUL, minimal memory. Primarily a perspective lens for evaluation in [[AI Agent Panels|panels]] or standalone feedback The distinction is a spectrum, not a binary. An agent starts light and becomes heavier as it accumulates memory and skills over time. ## Why Identity Matters Without a defined identity, agents produce generic, inconsistent results. With identity: - The agent stays in character across sessions - Boundaries prevent scope creep (an Editor doesn't try to write original content) - Decision frameworks make trade-offs explicit rather than implicit - Voice consistency makes multi-agent panels produce distinct, useful perspectives ## Connection to Separation of Concerns Agent identity is an application of [[Separation of Concerns]] to AI systems. Each agent owns a specific domain. Clear boundaries prevent agents from stepping on each other's toes. The result is a modular system where agents can be added, removed, or modified independently. ## References - ## Related - [[AI Agents]] - [[AI Agent Memory]] - [[AI Agent Panels]] - [[AI Agent Routing]] - [[AI Agent Skills]] - [[AI Assistant Architecture]] - [[Separation of Concerns]] - [[Receptionist AI Design Pattern]]