The 5 Files Every OpenClaw Operator Needs (And What Goes in Them)

The Files Are the System

OpenClaw agents don't run on prompts alone. The real infrastructure is the set of persistent files your agent reads at the start of every session, consults during execution, and writes to over time. Without these files, every session starts from zero. With them, your agent carries a working memory of your business, your preferences, and your operating rules — every single time.

There are five files that every operator needs. Here's what each one does and what breaks when it's missing.

1. SOUL.md — Identity and Operating Principles

SOUL.md is where your agent's character lives. It answers three questions: Who is this agent? What does it care about? How does it make decisions when things aren't clear?

What goes in it:

- The agent's operating name and role definition

- Core values (e.g., "prioritize accuracy over speed," "flag uncertainty rather than guess")

- Communication style and tone guidelines

- Decision-making framework for ambiguous situations

- Non-negotiable behaviors — things the agent always does or never does regardless of instructions

What breaks without it:

Without SOUL.md, your agent is a chameleon. Ask it two similar things on different days and you get two different versions. It'll match the energy of your prompt rather than maintaining a stable operating identity. Over time, this produces inconsistent outputs and means you're constantly re-calibrating.

SOUL.md is especially critical for client-facing work. If your agent is helping you run communications, you need it to sound like you — not like a generic AI assistant.

2. AGENTS.md — Autonomy Tiers and Security Rules

AGENTS.md defines what your agent can do on its own versus what it needs to ask you about. It's the security layer and the permission model for your entire workspace.

What goes in it:

- Tier 1 actions (full autonomy — no approval needed)

- Tier 2 actions (proceed unless stopped — notify first)

- Tier 3 actions (explicit approval always required)

- Failure protocols and escalation rules

- Operating hours and out-of-hours behavior

What breaks without it:

Without AGENTS.md, your agent defaults to either extreme: either it asks for approval on everything (useless), or it makes every decision on its own with no guardrails (dangerous). There's no middle ground, and there's no consistent rule for when to act versus when to stop.

AGENTS.md is also where you prevent the worst mistakes. Sending an email to the wrong person, publishing something early, accidentally deleting a file — all of these are prevented by clear tier definitions.

3. USER.md — Operator Profile

USER.md is the file that tells your agent who it's working for. It's the onboarding document you'd give a new hire, translated into a format the agent reads every session.

What goes in it:

- Your role, business model, and primary goals

- How you like to receive information (brief summaries vs. detailed breakdowns)

- Known preferences and pet peeves ("I hate bullet-pointed summaries, give me prose")

- Current priorities and focus areas

- Key relationships, clients, and projects

What breaks without it:

Without USER.md, your agent is working blind. It doesn't know if you're a solo freelancer or an agency with ten clients. It doesn't know whether you prefer to be briefed in the morning or at end of day. It doesn't know which clients are highest priority.

You end up re-explaining these basics at the start of every conversation. USER.md eliminates that entirely.

4. HEARTBEAT.md — Schedule and Recurring Routines

HEARTBEAT.md is your agent's calendar. It defines what happens when, what triggers recurring tasks, and how the agent structures its own work day.

What goes in it:

- Morning brief schedule and format

- Scheduled check-ins and status reviews

- Nightly consolidation and memory update routine

- Recurring task triggers (weekly reports, monthly reviews)

- Cron job definitions and timing

What breaks without it:

Without HEARTBEAT.md, your agent is entirely reactive. It waits for you to start every session. It has no rhythm, no schedule, no proactive behavior. You get a slightly smarter chatbot instead of an operator that shows up ready.

HEARTBEAT.md is also what converts your agent from a tool you use to a system that runs — the difference between pulling it in when you remember to versus having it already working when you check in.

5. MEMORY.md — Persistent Business Context

MEMORY.md is the running log of everything your agent knows about your business, your decisions, and your current state. It's the difference between an agent that has been working with you for months and one that met you five minutes ago.

What goes in it:

- Business context and history

- Decisions made and the reasoning behind them

- Active projects and current status

- Iron-law rules (things that never change)

- Open questions and pending items

What breaks without it:

Without MEMORY.md, every session is a first session. You rebuild context every time. You re-explain the same things. The agent makes suggestions that ignore previous decisions. You lose continuity.

Good MEMORY.md hygiene is also what prevents the agent from getting "stale" — outdated context that leads to bad suggestions. A well-maintained MEMORY.md is updated at the end of every session with what changed, what was decided, and what's coming next.

What a Complete Workspace Looks Like

A complete OpenClaw workspace has all five files populated, consistent, and current. The agent reads them at session start, consults them during execution, and updates the relevant ones before closing out.

A half-configured workspace has one or two of these files missing or stale. The symptoms: inconsistent agent behavior, repetitive context-setting, missed protocols, and decisions that contradict earlier ones.

The files don't have to be perfect from day one. They evolve. But you need all five present and at least minimally populated before your agent can operate like an operator rather than a chatbot.