name: multiagent description: "Manage multi-agent Telegram group collaboration for OpenClaw. Use when: (1) adding a new agent/bot to the team, (2) removing an agent, (3) listing current agents, (4) diagnosing multi-agent configuration issues, (5) clearing session caches, (6) syncing workspace team info. Triggers: 'multiagent', 'add agent', 'remove agent', 'agent doctor', 'agent list', 'clear sessions', 'multi-agent'." user_invocable: true

Multi-Agent Manager

Manage OpenClaw multi-agent Telegram group collaboration. Automates the error-prone process of adding/removing agents and keeps configurations consistent.

Operations

Script location: scripts/manage.sh

Adding an Agent

Prerequisites

1. Create a bot via BotFather (/newbot)
2. Disable privacy mode: /setprivacy -> select bot -> Disable
3. Add bot to the Telegram group
4. Note the bot token and username

Steps

bash scripts/manage.sh add-agent <id> <name> <role> <model> <bot_token> <bot_username> <group_chat_id>

Example:

bash scripts/manage.sh add-agent xiaoli 小丽 "数据分析师" "anthropic/claude-sonnet-4-6" \
  "1234567890:AAH..." "@xiaoli_data_bot" "-100XXXXXXXXXX"

This automatically:

1. Validates the agent ID is unique
2. Updates openclaw.json — 5 config sections:
   • agents.list[] — agent definition (model, workspace, mentionPatterns)
   • bindings[] — agentId <-> telegram accountId mapping
   • channels.telegram.accounts.{id} — bot token + group config
   • tools.agentToAgent.allow[] — adds new agent to global allow list
   • Each existing agent's subagents.allowAgents[] — bidirectional visibility
3. Creates workspace directory with AGENTS.md, SOUL.md, USER.md
4. Syncs AGENTS.md in all existing workspaces
5. Clears all session caches
6. Restarts the gateway

Post-add Checklist

• Verify bot privacy mode is disabled (can_read_all_group_messages: true)
• Remove and re-add bot to group if privacy was changed after joining
• Test: send @botname hello in group, verify the correct bot responds

Removing an Agent

bash scripts/manage.sh remove-agent <id>

Removes all configuration references, updates other agents' allow lists, syncs workspaces, clears sessions, restarts gateway. Does NOT delete the workspace directory (moved to .bak).

Listing Agents

bash scripts/manage.sh list-agents

Parses openclaw.json and displays each agent's ID, name, model, bot username, and binding status.

Diagnosing Issues (Doctor)

bash scripts/manage.sh doctor

Checks:

• Each agent has a corresponding binding
• Each agent has a telegram account config with bot token
• agentToAgent.allow includes all agent IDs
• Each agent's subagents.allowAgents includes all peers
• Workspace directories exist
• AGENTS.md uses accountId (not account)
• mentionPatterns are configured
• Group chat ID is consistent across accounts

Clearing Sessions

bash scripts/manage.sh clear-sessions [agent_id|all]

Deletes session cache files. Required after modifying AGENTS.md or SOUL.md — otherwise agents keep using cached instructions.

Session store locations: ~/.openclaw/agents/{id}/sessions/sessions.json

Updating Workspaces

bash scripts/manage.sh update-workspaces

Regenerates the team collaboration section in every agent's AGENTS.md with current team member list, ensuring all agents have consistent information.

Key Pitfalls

1. accountId not account — The message tool parameter is accountId. Writing account silently falls back to default (main agent's bot).

2. sessions_send never delivers to Telegram — It always uses webchat internally. Agents must use the message tool to post results to the group.

3. Session cache blocks config changes — After editing AGENTS.md or SOUL.md, clear sessions or changes won't take effect.

4. BotFather privacy mode — Must be Disabled. Change takes effect only after removing and re-adding bot to group.

5. Topic/forum mode adds complexity — Session keys get :topic:N suffixes. Use plain group mode unless you specifically need topics.

For detailed troubleshooting, read references/troubleshooting.md. For config structure details, read references/config-guide.md.

Workspace Templates

AGENTS.md Template

The template uses these placeholders (replaced by manage.sh):

Template content is embedded in scripts/manage.sh (function generate_agents_md).

SOUL.md Template

A minimal SOUL.md is generated with the agent's name and role. Edit it after creation to add personality details.

USER.md Template

A minimal USER.md pointing to the owner. Edit after creation.