claw-agents

name	claw-agents
description	Use when setting up or maintaining OpenClaw multi-agent routing, especially when the user wants `/claw-agents setup`, `/claw-agents add`, `/claw-agents status`, or `/claw-agents doctor` style workflows for WhatsApp, Telegram, Slack, or Discord.

Claw Agents

This skill packages OpenClaw multi-agent setup as a native /claw-agents ... workflow, with local scripts handling the actual file and CLI operations.

Trigger Intents

Use this skill when the user says things like:

• /claw-agents setup
• /claw-agents add work
• /claw-agents status
• /claw-agents doctor
• “配置 OpenClaw 多 agent”
• “给不同渠道绑定不同 agent”
• “检查 workspace / agentDir / bindings 对不对”

Command Mapping

Treat these as the canonical actions:

/claw-agents setup
/claw-agents add <agentId>
/claw-agents status
/claw-agents doctor

Backend mapping:

./scripts/openclaw-agents.sh setup
./scripts/openclaw-agents.sh add <agentId>
./scripts/openclaw-agents.sh status
./scripts/openclaw-agents.sh doctor

If the user also wants official channel provisioning, run:

./scripts/openclaw-agents.sh provision --channel <channel> --accounts <csv>

What Each Command Does

/claw-agents setup

Use for first-time setup. It will:

• collect channel and routing mode
• create one isolated workspace per agent
• create one isolated agentDir per agent
• create sessions, AGENTS.md, SOUL.md, USER.md, and auth-profiles.json
• write ~/.openclaw/openclaw.json
• optionally hand off to official openclaw channels ... provisioning

/claw-agents add <agentId>

Use to append one new agent to an existing config without hand-editing JSON.

/claw-agents status

Use to show the local config first, and when the openclaw CLI is installed, also show openclaw agents list --bindings.

/claw-agents doctor

Use to validate:

• workspace uniqueness
• agentDir uniqueness
• sessions existence
• auth-profiles.json
• binding order and binding completeness

Operating Rules

• Prefer the wrapper script over direct script calls.
• Keep one unique workspace and one unique agentDir per agent.
• Keep tools and sandbox inside each agent object.
• Put more specific bindings first: peer before guildId/teamId/accountId, then channel-wide matches, then default agent.
• Treat credentials as official CLI work. Do not synthesize tokens or login artifacts in the skill itself.

Post-Change Verification

After setup or add:

./scripts/openclaw-agents.sh doctor
openclaw agents list --bindings
openclaw channels status --probe

If probe fails or runtime state is stale:

openclaw gateway restart

Reference

• config-guide.md