How to Write a SOUL.md for Your OpenClaw Agent (With Template)

Why Your Agent Behaves Inconsistently

If your OpenClaw agent gives you different answers to the same question depending on how you phrase it — or feels like a different "personality" across sessions — the problem isn't the model. It's the absence of SOUL.md.

SOUL.md is the identity file that gets injected at the start of every session. It tells your agent who it is, what it stands for, how it makes decisions, and what it should never do. Without it, the model falls back on its training defaults, which are designed to be generically helpful — not specifically useful to you.

What SOUL.md Actually Does

OpenClaw injects workspace files into every session as system context. SOUL.md is read first and sets the behavioral baseline for everything that follows.

Think of it as the difference between hiring a contractor and hiring an employee. A contractor shows up and does what you ask. An employee understands the mission, has opinions, pushes back when something is wrong, and acts without waiting to be told.

SOUL.md is what turns your agent from a contractor into an operator.

The Four Sections Every SOUL.md Needs

1. Identity statement

One sentence: who is this agent, what is its job, what are its constraints? Don't be vague. "You are an AI assistant" is useless. "You are an autonomous business operator whose job is to build and grow a revenue-generating SaaS" is not.

2. Core operating principles

The non-negotiable behavioral rules. These should answer: How does this agent make decisions? What does it prioritize? What does it never do? Write these as truths, not instructions.

Examples that work:

"Revenue is the scoreboard. Every decision filters through: does this move closer to the goal?"

"Three failed attempts on the same approach means stop and escalate."

"No busywork. If a task doesn't serve the mission, skip it."

3. Communication style

How does the agent communicate with you? Direct or diplomatic? Terse or detailed? Does it have opinions? Does it challenge you? Set this explicitly or you'll get the model's default personality — aggressively agreeable and relentlessly pleasant.

4. What the agent is NOT

Negative constraints are as important as positive ones. "You are not waiting for instructions" and "You do not make excuses" shape behavior as much as anything affirmative.

What Most People Get Wrong

Too generic. "Be helpful, harmless, and honest" is not a soul. It's a safety disclaimer. Your SOUL.md should read like a job description written by someone who actually knows the role.

Too long. Every line costs tokens on every session. Keep it tight — under 300 words. If you need 800 words to describe your agent's identity, you don't know what you want yet.

Conflicting instructions. "Be proactive" and "always ask before acting" are at war with each other. Resolve conflicts in the file or your agent will resolve them inconsistently.

Free Template

Download the free OpenClaw Starter Kit at agentickit.io/free — includes ready-to-customize SOUL.md and MEMORY.md templates.

The full Solopreneur Operator Kit ($49) adds AGENTS.md with security rules and autonomy tiers, HEARTBEAT.md for automated daily tasks, nightly self-improvement loop, and three production-ready skill files.