SOUL.md is the 📝OpenClaw configuration file that defines an agent's personality — its temperament, core values, conversational voice, and behavioral guardrails. While 📝IDENTITY.md answers "Who are you?" SOUL.md answers "How should you behave?" It is the file that transforms a generic AI utility into a distinct personality with a consistent voice.

OpenClaw reads SOUL.md on startup and applies it as a foundational instruction layer to every interaction. Every response the agent generates is filtered through this file, making it the single most impactful configuration surface in the agent definition system.

Key Facts

File type: Markdown configuration file
Location: Workspace root (~/.openclaw/workspace/SOUL.md)
Loaded by: 📝BOOTSTRAP.md during agent initialization
Parent system: OpenClaw
Instruction priority: Foundational system-level prompt — higher priority than user messages
Recommended length: Under 2,000 words (loaded into every prompt; bloat wastes tokens and dilutes important rules)

How It Works

SOUL.md defines five dimensions of agent personality:

Tone and voice — how the agent communicates: formal vs. conversational, direct vs. diplomatic, concise vs. expansive
Opinions and stances — positions the agent holds on specific topics, giving it a genuine point of view rather than reflexive neutrality
Humor approach — whether and how the agent uses humor, wit, sarcasm, or levity
Boundaries and bluntness — how directly the agent pushes back on bad ideas, when it escalates, and where its candor limits sit
Brevity preferences — default response length, when to be terse vs. thorough

The official guidance is direct: "Short beats long. Sharp beats vague." An effective SOUL.md reads like "have a take," "skip filler," "call out bad ideas early" — not like corporate compliance language.

What Belongs in SOUL.md

Tone, voice, and style
Opinions and stances
Humor and personality
Behavioral guardrails (what the agent will and won't do)

What Does Not Belong

Life stories or changelogs (that's 📝MEMORY.md)
Operational role and job description (that's IDENTITY.md)
Security policies (that's 📝AGENTS.md)
Available tools and APIs (that's 📝TOOLS.md)
Corporate jargon or vague vibes without behavioral impact

Why It Matters

Personality in an agent is not cosmetic — it shapes how the agent drafts emails, summarizes information, explains complex topics, and navigates ambiguity. Two agents with identical tools and identical identities but different SOUL.md files will produce meaningfully different outputs because style is substance in communication.

In multi-agent systems like 📝Brian Bot's ecosystem, SOUL.md also serves as a coordination mechanism. Different agents can share the same organizational identity but express different personalities appropriate to their function — a research agent that is thorough and cautious, a social media agent that is punchy and opinionated, a customer support agent that is warm and patient.

The parallel in 📝MythOS is the Soul augmentation memo — one of four living memos (Soul, Style, Human, Memory) that shape how AI engages with a creator's library. The concept is the same: a structured, persistent personality layer that carries across sessions and contexts.

FAQ

What is the difference between SOUL.md and IDENTITY.md?

SOUL.md defines how the agent behaves — personality, tone, values, humor. IDENTITY.md defines what the agent is — name, role, boundaries, affiliations. An agent's soul is its personality; its identity is its job description.

How long should SOUL.md be?

Under 2,000 words. SOUL.md is loaded into every prompt, so every word consumes tokens on every interaction. The best SOUL.md files are tight — a few hundred words of sharp, specific instructions beat pages of vague personality description.

Can SOUL.md be updated while an agent is running?

Changes to SOUL.md take effect on the next agent restart or reinitialization. The file is read at startup, not watched for live changes.

How is SOUL.md different from a system prompt?

SOUL.md is a system prompt — but structured as a file rather than an inline string. This makes it version-controllable, templatable, and separable from the operational instructions that live in other files. The file-based approach means personality can be reviewed, diffed, and evolved like code.

Does MythOS have an equivalent to SOUL.md?

Yes. MythOS uses a Soul augmentation memo as part of its four-memo identity system (Soul, Style, Human, Memory), loaded via get_context at the start of every MCP session. The concept is structurally identical: a persistent personality layer that shapes every interaction.

📝OpenClaw — the open-source AI assistant platform this file belongs to
📝Markdown-File Agent Definition — the design pattern SOUL.md belongs to
📝IDENTITY.md — role, name, and operational boundaries
📝BOOTSTRAP.md — the initialization file that loads SOUL.md
📝MEMORY.md — long-term curated context
📝TOOLS.md — available APIs, functions, and integrations
📝Collaborative Augmentation — the broader framework for agent-human collaboration