Automates the full end-to-end flow for creating an OpenClaw Agent.
Activate when user says things like:
Ask the user:
qisi, code-review, brainstormoc_xxx (kind = group)ou_xxx (kind = direct)OPENCLAW_DIR=~/.openclaw
mkdir -p $OPENCLAW_DIR/workspace-{agent_id}
mkdir -p $OPENCLAW_DIR/workspace-{agent_id}/memory
mkdir -p $OPENCLAW_DIR/workspace-{agent_id}/skills
Create the following files in the workspace directory:
SOUL.md — Agent persona (see template below)USER.md — User info templateAGENTS.md — Session bootstrap instructionsHEARTBEAT.md — Heartbeat configTOOLS.md — Tool notesmemory/ — Daily memory folderIDENTITY.md — Agent identity definition (optional)BOOTSTRAP.md — First-run bootstrap (optional)AGENTS.md content (copied from ~/.openclaw/workspace/AGENTS.md):
# AGENTS.md - Your Workspace
If `BOOTSTRAP.md` exists, that's your birth certificate. Follow it, figure out who you are, then delete it.
## Every Session
Before doing anything else:
1. Read `SOUL.md` — this is who you are
2. Read `USER.md` — this is who you're helping
3. Read `memory/YYYY-MM-DD.md` (today + yesterday) for recent context
4. **If in MAIN SESSION** (direct chat with your human): Also read `MEMORY.md`
## Memory
You wake up fresh each session. These files are your continuity:
- **Daily notes:** `memory/YYYY-MM-DD.md` — raw logs of what happened
- **Long-term:** `MEMORY.md` — your curated memories
## Safety
- Don't exfiltrate private data
- When in doubt, ask.
HEARTBEAT.md content:
# HEARTBEAT.md
# Keep this file empty to skip heartbeat API calls.
TOOLS.md content:
# TOOLS.md - Local Notes
Notes specific to this agent's setup.
OPENCLAW_DIR=~/.openclaw
mkdir -p $OPENCLAW_DIR/agents/{agent_id}/agent
mkdir -p $OPENCLAW_DIR/agents/{agent_id}/sessions
OPENCLAW_DIR=~/.openclaw
cp $OPENCLAW_DIR/agents/main/agent/models.json $OPENCLAW_DIR/agents/{agent_id}/agent/
cp $OPENCLAW_DIR/agents/main/agent/auth-profiles.json $OPENCLAW_DIR/agents/{agent_id}/agent/
If the main agent config files don't exist, warn the user and ask them to provide models.json and auth-profiles.json manually.
Back up first:
cp ~/.openclaw/openclaw.json ~/.openclaw/openclaw.json.bak
Read the current config and add the new agent to agents.list:
{
"id": "{agent_id}",
"name": "{agent_name}",
"workspace": "~/.openclaw/workspace-{agent_id}",
"agentDir": "~/.openclaw/agents/{agent_id}/agent"
}
Add bindings if provided. For group chat:
{
"agentId": "{agent_id}",
"match": {
"channel": "feishu",
"peer": {
"kind": "group",
"id": "{group_chat_id}"
}
}
}
For direct chat:
{
"agentId": "{agent_id}",
"match": {
"channel": "feishu",
"peer": {
"kind": "direct",
"id": "{user_id}"
}
}
}
If the agent needs specific skills:
cp -r ~/.openclaw/skills/{skill-name} ~/.openclaw/workspace-{agent_id}/skills/
openclaw gateway restart
Tell the user:
# SOUL.md - {Agent Name}
_{One-line purpose}_
## Core Identity
**You are "{Agent Name}"** — {detailed description}
## Principles
- {principle 1}
- {principle 2}
- {principle 3}
## Style
- {style description}
---
_{Mission statement}_
After creation, the layout should be:
~/.openclaw/
├── agents/
│ └── {agent_id}/
│ ├── agent/
│ │ ├── auth-profiles.json
│ │ └── models.json
│ └── sessions/
└── workspace-{agent_id}/
├── AGENTS.md
├── BOOTSTRAP.md (optional)
├── HEARTBEAT.md
├── IDENTITY.md (optional)
├── SOUL.md
├── TOOLS.md
├── USER.md
├── memory/
└── skills/ (optional)
workspace-brainstormopenclaw.json before modifyingmodels.json and auth-profiles.json to the agents directoryscripts/agent_creator.py to automate the config and directory creation共 1 个版本