← Back to Skills Marketplace
eggyrooch-blip

Openclaw Team Builder

by eggyrooch-blip · GitHub ↗ · v3.6.1
darwinlinux ⚠ suspicious
741
Downloads
1
Stars
6
Active Installs
10
Versions
Install in OpenClaw
/install openclaw-team-builder
Description
Manage OpenClaw agent teams — add agents, deploy templates, health check, auto-fix, view org tree, rollback, goal-driven team suggestion. Use when user menti...
README (SKILL.md)

OpenClaw Team Builder

Manage your AI agent team: add agents to any position in the org tree, deploy templates, run health checks, auto-fix issues, and rollback changes.

Setup

The script is at ~/.openclaw/skills/team-builder/scripts/team-builder.sh. It requires openclaw CLI and python3.

TB="bash ~/.openclaw/skills/team-builder/scripts/team-builder.sh"

All examples below use $TB as shorthand.

View Org Tree

# Human-readable tree
$TB --tree

# JSON output (for parsing)
$TB --tree --json

JSON output example:

{
  "main": {"name":"软软","emoji":"😘","role":"director","reports_to":null,"manages":["xingzheng"],"description":""},
  "xingzheng": {"name":"xingzheng","emoji":"🤖","role":"worker","reports_to":"main","manages":[],"description":""}
}

Add an Agent

# Full CLI mode (no prompts)
$TB --add \
  --id finance-lead \
  --name "小财-财务助手" \
  --emoji "💰" \
  --role "负责报销审核、预算管理、财务报表" \
  --parent main \
  --soul auto \
  --yes

# Use a built-in role template (auto-fills role description)
$TB --add \
  --id caiwu \
  --soul template:caiwu \
  --parent main \
  --yes

# With model override and feishu config
$TB --add \
  --id translator \
  --name "翻译官" \
  --emoji "🌐" \
  --role "负责中英文翻译" \
  --parent main \
  --model anthropic/claude-sonnet-4-6 \
  --feishu-app-id cli_xxx \
  --feishu-secret yyy \
  --yes

Parameters

Parameter Required Default Description
--id Yes - Agent ID (english, kebab-case)
--name No same as id Display name
--emoji No 🤖 Agent emoji
--role No 通用AI助手 Role description for SOUL.md
--parent No main Parent agent ID in org tree
--soul No auto auto, template:\x3Ckey>, or skip
--model No inherit Model override
--feishu-app-id No - Feishu bot App ID
--feishu-secret No - Feishu bot App Secret
--yes No false Skip confirmation prompts

Soul modes

  • auto — Generate SOUL.md from role description + org relationships
  • template:\x3Ckey> — Use built-in template (also sets name/emoji/role automatically)
  • skip — Keep OpenClaw default template

Goal-Driven Team Suggestion

Given a business goal, recommend the best team configuration from built-in templates.

# Get recommendation (human-readable)
$TB --suggest --goal "电商平台运营"

# JSON output (for AI Agent parsing)
$TB --suggest --goal "电商平台运营" --json

JSON output example:

{
  "goal": "电商平台运营",
  "matched_scenario": "ecommerce",
  "scenario_name": "电商团队",
  "recommended_agents": [
    {"id": "kefu", "template": "kefu", "reason": "处理客户咨询和售后"},
    {"id": "yunying", "template": "yunying", "reason": "运营数据分析和活动策划"}
  ],
  "deploy_commands": ["$TB --add --id kefu --soul template:kefu --parent main --yes", "..."],
  "total_agents": 4
}

Supported scenarios: ecommerce(电商), content(内容创作), devteam(研发), startup(创业), consulting(专业服务), solo(超级个体).

AI Agent Workflow: Creating a New Agent

When the user wants to add a new agent, follow these exact steps in order.

CRITICAL: Collect ALL info in ONE message. Do NOT use multi-step Q&A — session context may be lost between messages.

Step 1: Ask ONE compound question (all in a single message):

"好的!帮你创建新助手,请一次性告诉我: 1️⃣ 名字和 emoji(比如:小财 💰) 2️⃣ 核心任务(一句话,比如:负责报销和财务) 3️⃣ 自主级别:建议/执行/全自动 4️⃣ 语调:专业/亲切/简洁

比如:'小财 💰,负责报销和预算,全自动,简洁'"

If the user gives a short answer (e.g. "财务助手"), infer reasonable defaults:

  • name: 小财 emoji: 💰 role: 财务管理 level: 执行 tone: 专业

If the user gives partial info, fill in sensible defaults and proceed. Do NOT ask follow-up questions.

Step 2: Immediately construct and execute (do NOT use openclaw agents add directly):

$TB --add \
  --id \x3Ckebab-case-id> \
  --name "\x3Cname>" \
  --emoji "\x3Cemoji>" \
  --role "核心任务:\x3Ctask>。自主级别:\x3Clevel>。语调:\x3Ctone>" \
  --parent main \
  --soul auto \
  --yes

Then restart gateway:

openclaw gateway restart

The script automatically: creates agent, generates SOUL.md, configures agentToAgent, and binds to ALL enabled channels (Telegram, Discord, Feishu, WeChat, iMessage).

Step 3: Confirm to user (in the SAME response as execution):

"✅ \x3Cname> 已创建并加入团队!已自动绑定所有可用渠道。"

Then show the tree:

$TB --tree

Step 4: Channel configuration (in the SAME response):

After creation, check which channels the agent is missing:

$TB --channels --agent \x3Cid> --json

Then guide the user based on missing_channels:

"📡 \x3Cname> 缺少以下渠道:\x3Cmissing_channels>

要配置独立的 bot 吗?

  • Telegram:请到 @BotFather 创建 bot,把 token 发给我
  • Discord:请到 Developer Portal 创建 bot,把 token 发给我
  • 飞书:请提供 App ID 和 App Secret
  • 不需要的话可以跳过。"

When user provides credentials:

# Telegram
$TB --channels --agent \x3Cid> --channel telegram --token \x3Ctoken> --yes

# Discord
$TB --channels --agent \x3Cid> --channel discord --token \x3Ctoken> --yes

# Feishu
$TB --channels --agent \x3Cid> --channel feishu --feishu-app-id \x3Cid> --feishu-secret \x3Cs> --yes

Then restart gateway:

openclaw gateway restart

IMPORTANT:

  • Always use $TB --add, never directly call openclaw agents add
  • Never split into multiple Q&A rounds — collect everything in one message
  • If user says something like "加个客服" — use the template: $TB --add --id kefu --soul template:kefu --parent main --yes
  • After --add, always show channel binding status and offer feishu config

Deploy Solo Template

Deploy the "Super Individual" team: 4 specialist agents under main (dev, design, content, data).

# Deploy with defaults
$TB --solo --yes

# Deploy with specific model
$TB --solo --model anthropic/claude-sonnet-4-6 --yes

# Skip SOUL generation
$TB --solo --soul skip --yes

Already-existing agent IDs are automatically skipped.

List Role Templates

# Human-readable
$TB --templates

# JSON
$TB --templates --json

Available templates: xingzheng(行政), caiwu(财务), hr(人力), kefu(客服), yunying(运营), falv(法务), neirong(内容), shuju(数据), jishu(技术)

Channel Management

View channel status, agent bindings, and add new bots for agents.

# List all channels and bindings
$TB --channels --json

# Filter by agent
$TB --channels --agent kefu --json

# Add a Telegram bot for an agent
$TB --channels --agent kefu --channel telegram --token \x3Cbot-token> --yes

# Add a Discord bot for an agent
$TB --channels --agent kefu --channel discord --token \x3Cbot-token> --yes

# Add a Feishu app for an agent
$TB --channels --agent kefu --channel feishu --feishu-app-id cli_xxx --feishu-secret yyy --yes

JSON output example:

{
  "channels": {
    "telegram": {"enabled": true, "type": "shared", "accounts": []},
    "feishu": {"enabled": true, "type": "per-agent", "accounts": ["main"]}
  },
  "agents": [
    {
      "id": "main", "name": "软软",
      "bindings": [{"channel": "telegram", "accountId": ""}],
      "missing_channels": ["feishu"]
    }
  ]
}

Channel Model

  • Each Agent can have its own bot (Telegram/Discord/Feishu etc.)
  • A shared bot can only bind to ONE agent
  • To have multiple agents on the same channel, each needs its own bot

How to create bots (guide the user):

  • Telegram: Talk to @BotFather on Telegram → /newbot → get token
  • Discord: Discord Developer Portal → New Application → Bot → Copy Token
  • Feishu: 飞书开放平台 → 创建应用 → 获取 App ID + App Secret

AI Agent Workflow: Channel Configuration

After creating an agent, check channel status and guide the user:

$TB --channels --agent \x3Cid> --json

If the agent has missing channels, ask the user:

"📡 \x3Cname> 目前缺少以下渠道绑定:\x3Cmissing_channels>

要为 \x3Cname> 配置独立的 bot 吗?

  • Telegram:请先到 @BotFather 创建 bot,把 token 发给我
  • Discord:请到 Developer Portal 创建 bot,把 token 发给我
  • 飞书:请提供 App ID 和 App Secret

也可以跳过,之后随时用 --channels 配置。"

When the user provides credentials, execute:

# Telegram
$TB --channels --agent \x3Cid> --channel telegram --token \x3Ctoken> --yes

# Discord
$TB --channels --agent \x3Cid> --channel discord --token \x3Ctoken> --yes

# Feishu
$TB --channels --agent \x3Cid> --channel feishu --feishu-app-id \x3Cid> --feishu-secret \x3Cs> --yes

Then restart gateway and confirm:

openclaw gateway restart

Health Check

Scans for: gateway status, agentToAgent config, allow list completeness, SOUL.md presence, binding completeness, hierarchy file.

# Human-readable
$TB --checkup

# JSON
$TB --checkup --json

JSON output example:

{
  "checks": [
    {"name": "gateway", "status": "ok", "detail": null},
    {"name": "a2a_allow", "status": "warn", "detail": "缺失: xingzheng"}
  ],
  "issues": 1
}

Auto-Fix

Automatically fixes issues found by health check: restart gateway, enable agentToAgent, fill allow list, generate missing SOUL.md, add bindings, init hierarchy.

# Auto-fix (with confirmation)
$TB --fix

# Auto-fix (no prompts)
$TB --fix --yes

Team Status

Full overview: org tree, agent list, a2a config, bindings, SOUL.md line counts, backup count.

# Human-readable
$TB --status

# JSON (comprehensive)
$TB --status --json

Rollback

Every operation creates a backup. Rollback restores config, hierarchy, and SOUL.md files, and deletes agents created in that operation.

# Interactive rollback (pick from list)
$TB --rollback

# Rollback to most recent backup
$TB --rollback --index 1 --yes

Workflow Examples

"帮我加一个财务助手"

$TB --add --id caiwu --soul template:caiwu --parent main --yes

"看看团队现在什么状态"

$TB --status --json

"团队体检一下,有问题就修"

$TB --checkup --json
# If issues > 0:
$TB --fix --yes

"刚才加错了,撤回"

$TB --rollback --index 1 --yes

Notes

  • All write operations auto-backup before executing (max 5 backups kept)
  • Run openclaw gateway restart after changes to apply
  • The script preserves existing agents — never overwrites
  • TUI mode (no args) provides a full interactive menu for human use
Usage Guidance
This skill appears to implement a legitimate team-management wrapper for OpenClaw, but it asks the agent to collect channel credentials in a single message and then immediately run automated binding and restart commands. Before installing or invoking: 1) prefer running the provided script locally from your shell (not pasting secrets into chat). 2) If you must provide tokens, create least-privilege tokens or throwaway/test accounts. 3) Inspect where the script writes credentials (openclaw config/backups) and confirm you are comfortable with that storage. 4) Back up ~/.openclaw/openclaw.json and team-hierarchy.json before using automated add/fix/rollback. 5) If you will let the agent invoke the skill autonomously, be cautious — the workflow intentionally discourages multi-step confirmation, so limit the skill's use to trusted operators or run it interactively yourself.
Capability Analysis
Type: OpenClaw Skill Name: openclaw-team-builder Version: 3.6.1 The OpenClaw Team Builder skill is a legitimate administrative tool designed to manage AI agent hierarchies and configurations. It provides functionality for adding agents, visualizing team structures, performing health checks, and configuring communication channels (Telegram, Discord, Feishu) using the official OpenClaw CLI. The core logic in scripts/team-builder.sh includes robust backup and rollback mechanisms, ensuring configuration safety. While the skill handles sensitive bot tokens and modifies the openclaw.json config, its behavior is transparently aligned with its stated purpose, and the SKILL.md instructions provide a structured, user-confirmed workflow without any indicators of data exfiltration or unauthorized remote execution.
Capability Assessment
Purpose & Capability
Name/description align with required binaries (openclaw, python3) and the included shell script implements team/hierarchy, add/suggest/rollback and channel binding. Requiring openclaw and python3 is expected for this purpose.
Instruction Scope
SKILL.md explicitly instructs the agent to collect ALL agent configuration and (when needed) channel credentials in ONE message, then immediately run the wrapper ($TB --add ...) and restart the gateway. That workflow pushes users to paste sensitive tokens into chat/one-shot input and prevents multi-step confirmation — increases risk of unintended credential disclosure and reduces opportunity for user review.
Install Mechanism
This is instruction-only plus an included shell script; there is no remote download/install step. No high-risk install URLs or package installs are present in the provided files.
Credentials
The skill declares no required environment variables, but it expects users to provide channel credentials (Telegram/Discord/Feishu/WeChat/iMessage) via CLI arguments or by pasting them into chat. Requesting those tokens is coherent with channel binding, but collecting them via a single chat message and then invoking automated binding/restarts is sensitive and should be justified and handled carefully (e.g., use minimal-scope tokens and prefer local CLI input rather than sending tokens through chat).
Persistence & Privilege
The script reads and writes OpenClaw config and team-hierarchy files, creates backups, and issues 'openclaw gateway restart' — these are expected for managing agents but are privileged operations. always:false — it does not force-install globally, which is appropriate; however, autonomous invocation combined with the scripted restart/bind behavior increases blast radius if misused.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install openclaw-team-builder
  3. After installation, invoke the skill by name or use /openclaw-team-builder
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v3.6.1
- Internal shell script improvements in team-builder.sh for stability and reliability - No changes to features, user-facing CLI, or documentation - All workflows, parameters, and output formats remain the same
v3.6.0
openclaw-team-builder 3.6.0 - Channel management greatly improved: new syntax for adding Telegram/Discord/Feishu bots per agent, including with tokens. - Agent channel binding status now shows missing channels; guided instructions for adding missing bots integrated. - Documentation updated: clearer, step-by-step channel bot setup for agents; reflects new capabilities in channel management. - `$TB --channels --agent <id>` now supports adding new bot/app credentials and shows precise binding state. - Backward compatibility preserved for existing org tree and agent management commands.
v3.5.1
openclaw-team-builder 3.5.1 - Added new triggers ("渠道", "channel", "绑定", "飞书配置") for improved channel management detection. - Introduced dedicated channel management commands: - `$TB --channels` lists all channels, bindings, and status (now in both human-readable and JSON formats). - Support for configuring Feishu appId and appSecret per agent via CLI flags. - Updated documentation to explain the distinction between shared channels (e.g., Telegram) and per-agent channels (Feishu), including configuration workflow and JSON output examples. - No breaking changes; all existing workflows continue to function as before.
v3.5.0
- Added rollback shortcut command (`$TB --r`) for quick undo of the last team modification. - Updated documentation with rollback workflow example using the new shortcut. - No changes to functionality or parameters besides the new rollback alias. - All existing features and usage remain the same.
v3.4.3
- Added a new step to the agent creation workflow: after creating an agent, show channel binding status and offer Feishu bot configuration to the user. - Updated documentation to explain that after agent creation, the system will prompt for Feishu appId/appSecret if needed, while other channels are automatically bound. - Clarified that after using `$TB --add`, the channel configuration and Feishu setup should always be included in the same response.
v3.4.2
**New workflow: all agent info must be collected in a single message (no multi-step Q&A).** - Updated the AI Agent Workflow section: now requires users to provide all details (name, emoji, task, level, tone) in one message when creating a new agent. - Added guidance for defaulting/filling in missing info if the user's reply is partial. - Updated confirmation and execution steps to reflect the new single-message flow. - Emphasized not to use multi-step questioning in the workflow. - Clarified execution and confirmation should happen together in a single response.
v3.4.1
- Improved robustness for determining the script's own directory, enhancing compatibility across more bash environments. - No changes to user-facing behavior or commands.
v3.4.0
**Improved agent creation workflow requiring a structured user interview and new process guidance.** - Updated agent creation flow: AI agent now must interview the user with 4 specific questions before adding a new agent. - Added explicit step-by-step workflow for AI agents when creating new agents (interview, command construction, confirmation). - Clarified that `$TB --add` must be used for agent creation and raw CLI is not sufficient. - Script now automatically binds new agents to all enabled channels (Telegram, Discord, Feishu, etc.). - SKILL.md documentation updated to reflect new exact workflow and instructions for both agents and users.
v3.3.0
**openclaw-team-builder 3.3.0 changelog** - Added goal-driven team suggestion: `$TB --suggest --goal "your goal"` recommends an agent team for a business scenario. - Expanded triggers and description to support team recommendation use cases. - SKILL and README updated with instructions and recommended agent "interview" flow. - No breaking changes to existing commands.
v3.2.0
**Enhanced team management and automation for OpenClaw agent orgs** - Add agents with flexible role, model, and template options, supporting Feishu integration. - New commands for viewing, deploying, and customizing organizational trees and templates. - Health check and auto-fix features scan for and repair common configuration issues. - Full status overview and rollback system enables safer changes and quick restoration. - Improved documentation with CLI examples for common AI agent team workflows.
Metadata
Slug openclaw-team-builder
Version 3.6.1
License
All-time Installs 6
Active Installs 6
Total Versions 10
Frequently Asked Questions

What is Openclaw Team Builder?

Manage OpenClaw agent teams — add agents, deploy templates, health check, auto-fix, view org tree, rollback, goal-driven team suggestion. Use when user menti... It is an AI Agent Skill for Claude Code / OpenClaw, with 741 downloads so far.

How do I install Openclaw Team Builder?

Run "/install openclaw-team-builder" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Openclaw Team Builder free?

Yes, Openclaw Team Builder is completely free (open-source). You can download, install and use it at no cost.

Which platforms does Openclaw Team Builder support?

Openclaw Team Builder is cross-platform and runs anywhere OpenClaw / Claude Code is available (darwin, linux).

Who created Openclaw Team Builder?

It is built and maintained by eggyrooch-blip (@eggyrooch-blip); the current version is v3.6.1.

💬 Comments