5.3 KiB
OpenClaw Agent Template
A starter kit for creating new OpenClaw agents. Get a fully configured agent running in under 5 minutes.
Quick Start (3 Steps)
Step 1 — Clone this template
# Copy the template files to your agent's workspace directory
cp -r openclaw-agent-template/ ~/.openclaw/projects/YOUR_AGENT_NAME/
Or if you're setting up inside an existing project:
# Copy just the agent config files into your project
cp openclaw-agent-template/{SOUL.md,IDENTITY.md,USER.md,AGENTS.md,TOOLS.md,MEMORY.md} your-project/
cp -r openclaw-agent-template/memory/ your-project/memory/
Step 2 — Fill in the placeholders
Open each file and replace the [PLACEHOLDER] values with your agent's actual details:
| File | What to fill in |
|---|---|
IDENTITY.md |
Agent name, emoji, creature, vibe |
USER.md |
Your name, pronouns, timezone, preferences |
SOUL.md |
Personality description, workspace name |
AGENTS.md |
Workspace name, agent role, capabilities |
TOOLS.md |
Workspace-specific tools and restrictions |
MEMORY.md |
Leave empty — the agent will populate this |
memory/CURRENT_STATE.md |
Leave empty — the agent will populate this |
Search for [ to find all placeholders across files:
grep -r "\[" . --include="*.md" -l
Step 3 — Restart the gateway
openclaw gateway restart
# or via the OpenClaw app: Settings -> Gateway -> Restart
Your agent is ready. Open a new chat and it will introduce itself.
File Reference
SOUL.md — Behavioral DNA
Defines how the agent communicates and recovers from context loss. Contains:
- Message prefix format — the emoji + name pattern used in every message
- Core truths — non-negotiable facts the agent holds about itself
- Vibe — personality description that shapes tone
- Continuity recovery protocol — step-by-step instructions for re-orienting after a context gap
Edit this when: you want to change the agent's personality, communication style, or recovery behavior.
IDENTITY.md — Who the Agent Is
Defines the agent's name, emoji, creature, role, and specializations.
Edit this when: the agent takes on new responsibilities or you want to update its self-description.
USER.md — Who the Agent Serves
Tells the agent about the person it's working with — expertise level, preferences, timezone, and working style. This helps the agent calibrate its responses.
Edit this when: your working style changes, you want to add/remove preferences, or a new person starts working with the agent.
AGENTS.md — Workspace Map
Describes the workspace and all agents operating within it. Follows the standard OpenClaw AGENTS.md format. Useful when multiple agents share a workspace.
Edit this when: you add a new agent, change workspace rules, or update shared resources.
TOOLS.md — What the Agent Can Do
Documents tools available to the agent: built-in Claude Code tools, workspace-specific CLIs, MCP servers, and APIs. Also defines restricted actions that require user approval.
Edit this when: you add new tools to the workspace, change permissions, or want to restrict certain actions.
MEMORY.md — Long-Term Memory
Accumulates important decisions, context, and learnings across sessions. The agent reads this when starting a new session or recovering context.
Edit this when: you want to manually add context or remove outdated entries. The agent will also write to this file automatically.
memory/CURRENT_STATE.md — Right Now
Tracks what's in flight, what's blocked, what was recently finished, and what's planned next. The agent updates this at the end of every session.
Edit this when: you want to manually set the agent's initial task list or correct its understanding of current state.
FAQ
Q: Do I need all these files?
A: SOUL.md, IDENTITY.md, and USER.md are the minimum for a working agent. The others improve reliability and multi-session continuity.
Q: What are the [PLACEHOLDER] values?
A: Bracketed ALL_CAPS values are placeholders you should replace. <!-- comments --> explain what goes there. Search for [ to find them all.
Q: Can I have multiple agents in one workspace?
A: Yes. Add each agent to AGENTS.md and give each its own config directory. They'll share the workspace rules but have separate identities and memories.
Q: How does the agent remember things between sessions?
A: It reads MEMORY.md and memory/CURRENT_STATE.md at the start of each session. For this to work reliably, the agent needs to update CURRENT_STATE.md before each session ends.
Q: What if the agent seems confused or off-character?
A: Trigger continuity recovery manually by saying: "Re-read your SOUL.md and IDENTITY.md and re-orient." The agent will follow the recovery protocol defined in SOUL.md.
Q: Can I add custom tools?
A: Yes — document them in TOOLS.md. For MCP servers, configure them in Claude Code settings and reference them in TOOLS.md so the agent knows they exist and how to use them.
Q: How do I update the agent's personality?
A: Edit the ## Vibe section in SOUL.md. Changes take effect in the next session.
Template Version
v1.0.0 — Initial release
License
This template is provided as-is for use with OpenClaw. Modify freely.