If you’re referring to an **OpenClaw-style AI agent**, `SOUL.md` is one of several files that collectively define the agent. These files give it personality, instructions, knowledge about its user, memory, and capabilities.

## Core agent files

| File           | Purpose                                                                                                                                               |
| -------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------- |
| `SOUL.md`      | Defines the agent’s personality, values, tone, character, and behavioral boundaries.                                                                  |
| `IDENTITY.md`  | Defines who the agent is: name, role, purpose, background, and possibly its avatar or theme.                                                          |
| `USER.md`      | Describes the person the agent serves: preferences, goals, communication style, responsibilities, and relevant context.                               |
| `AGENTS.md`    | The agent’s operating manual—rules, workflows, permissions, safety requirements, and instructions for completing work.                                |
| `TOOLS.md`     | Notes about the agent’s available tools and environment: servers, applications, devices, APIs, paths, and usage conventions.                          |
| `MEMORY.md`    | Curated long-term memory containing durable facts, decisions, preferences, projects, and unresolved matters.                                          |
| `HEARTBEAT.md` | Instructions for recurring or proactive checks the agent should perform when awakened on a schedule.                                                  |
| `BOOTSTRAP.md` | First-run setup instructions used to establish the agent’s identity, personality, and relationship with its user. It is normally deleted after setup. |

These are the files OpenClaw injects into the agent’s working context when present. [OpenClaw context documentation](https://docs.openclaw.ai/concepts/context)

## Memory directory

```text
memory/
├── 2026-07-16.md
├── 2026-07-17.md
└── ...
```

Daily memory files record recent activity, such as:

* What happened that day
* Decisions made
* Work completed
* New preferences learned
* Commitments and follow-ups
* Problems still unresolved

A useful distinction is:

* `memory/YYYY-MM-DD.md` = daily journal or working history
* `MEMORY.md` = important long-term knowledge

The agent usually reads today’s file, yesterday’s file, and `MEMORY.md` when beginning a session. [OpenClaw default agent instructions](https://docs.openclaw.ai/reference/AGENTS.default)

## Skills

```text
skills/
├── research/
│   └── SKILL.md
├── email-management/
│   └── SKILL.md
└── business-planning/
    └── SKILL.md
```

Each `SKILL.md` teaches the agent how to perform a specialized task. A skill might include:

* When it should be used
* Step-by-step workflow
* Required tools
* Quality standards
* Safety precautions
* Output format
* Examples

The difference between `TOOLS.md` and `SKILL.md` is:

* `TOOLS.md` says **what resources are available and what is unique about the environment**.
* `SKILL.md` explains **how to perform a particular kind of work**.

## Supporting configuration

Outside the workspace, an agent system may also have:

* Model/provider configuration
* API credentials
* Channel connections, such as Slack or Telegram
* Permissions and approval policies
* Scheduled jobs
* Plugins
* Session transcripts
* Sandboxing rules
* Routing configuration for multiple agents

These technically help run the agent, but they are not part of its personality and knowledge workspace. OpenClaw keeps configuration, credentials, and session data separate from the agent’s workspace. [OpenClaw workspace documentation](https://docs.openclaw.ai/concepts/agent-workspace)

## Recommended structure

```text
ceo-agent/
├── AGENTS.md
├── SOUL.md
├── IDENTITY.md
├── USER.md
├── TOOLS.md
├── MEMORY.md
├── HEARTBEAT.md
│
├── memory/
│   ├── 2026-07-16.md
│   └── 2026-07-17.md
│
├── skills/
│   ├── strategic-planning/
│   │   └── SKILL.md
│   ├── financial-analysis/
│   │   └── SKILL.md
│   ├── leadership-coaching/
│   │   └── SKILL.md
│   └── executive-communications/
│       └── SKILL.md
│
├── knowledge/
│   ├── company-profile.md
│   ├── business-plan.md
│   ├── leadership-principles.md
│   └── current-priorities.md
│
└── projects/
    ├── active-projects.md
    └── decision-log.md
```

The simplest way to understand the system is:

* **Soul:** How I behave
* **Identity:** Who I am
* **User:** Who I serve
* **Agents:** How I operate
* **Tools:** What I can use
* **Skills:** What I know how to do
* **Memory:** What I remember
* **Heartbeat:** What I check proactively
* **Knowledge:** What I know about the business
* **Projects:** What I am currently responsible for