← 返回 Skills 市场
413
总下载
2
收藏
2
当前安装
1
版本数
在 OpenClaw 中安装
/install multi-agent-builder
功能描述
Build a reusable multi-agent team in OpenClaw from a user goal (e.g., "create a product-engineering team", "build a marketing ops team"). Use when the user w...
安全使用建议
This skill appears to do what it claims (create agent config, role files, and a report), but it also contains behaviors you should review before running: 1) Inspect scripts locally: materialize_team.mjs and create_team.mjs will modify /root/.openclaw/openclaw.json and create /root/.openclaw/workspace-<team> — confirm you want those changes and check the .bak file after a run. 2) Confirm binding behavior: the scripts will add a binding if --account-id is provided; do not pass an account/token unless you explicitly want an automatic binding. 3) Permission/profile review: materialize_team.mjs assigns tools: {profile:'full'} to agents—verify these profiles are acceptable and tighten them if you want least-privilege. 4) Skill installs: the playbook describes automatic skill installation and scanning (skillhub/clawhub/skill-vetter), but no installer is implemented here — if you plan to automate installs, require manual approval per item and validate scanners. 5) Run in a sandbox or with a dry-run first: execute validate_team.mjs against a copy of your openclaw.json or a test environment to see what would change. 6) If you need higher assurance, ask the author to: (a) provide an explicit dry-run flag, (b) avoid auto-binding in materialize (require interactive consent), and (c) stop assigning a 'full' tool profile by default. Finally, do not grant any channel tokens or admin credentials until you have reviewed and approved the exact binding steps and created backups of your config.
功能分析
Type: OpenClaw Skill
Name: multi-agent-builder
Version: 1.0.0
This bundle acts as a high-privilege administrative utility designed to automate the creation and configuration of multi-agent teams. The core script `scripts/materialize_team.mjs` performs sensitive operations, including direct modification of the global `/root/.openclaw/openclaw.json` configuration and the creation of agent workspaces with broad permissions (e.g., `exec` and `process` tools defined in `references/capability-matrix.md`). While the bundle includes extensive security-oriented documentation and instructions for 'skill-vetting' and 'least-privilege' (e.g., `references/provisioning-playbook.md`), the inherent capability to auto-install third-party skills and programmatically redefine system-wide agent boundaries represents a significant risk for privilege escalation or supply chain compromise if the agent is targeted by prompt injection.
能力评估
Purpose & Capability
Name/description (build multi-agent teams) aligns with the included scripts and many reference documents: the package materializes agents, creates role files, updates openclaw.json, and emits a report. However some requested behaviors are broader than necessary for the stated goal: materialize_team.mjs unconditionally sets each agent's tools to a 'full' profile and populates model primary+fallback lists, which is more permissive than the permission-profiles guidance in the references. The provisioning playbook also describes automatic installation of required/optional skills (skillhub/clawhub + scanners), yet no implementation for that install pipeline is present in the shipped scripts—so the spec and actual code diverge.
Instruction Scope
SKILL.md explicitly requires confirmation for channel/bot credential binding and other irreversible external effects, but the playbook and scripts allow materialization and (if accountId is provided) will add bindings and write openclaw.json automatically. create_team.mjs -> materialize_team.mjs writes /root/.openclaw/openclaw.json and creates /root/.openclaw/workspace-<team>/shared/ without a user prompt. The provisioning-playbook also prescribes automatic skill installs without per-item confirmation. These are scope inconsistencies: the instructions both require confirmation and describe automatic changes, giving the agent capability to perform system-level config changes contingent on inputs (accountId) that the user may not expect.
Install Mechanism
There is no installer that downloads/extracts remote binaries — the skill is instruction+local scripts only, which lowers supply-chain risk. The scripts write files under /root/.openclaw, create workspaces, and atomically replace openclaw.json (they create a .bak). That means the runtime will persist changes to system config and disk. No network downloads are present in the code, but the SKILL.md references searching/installing skills from 'skillhub'/'clawhub' and running 'skill-vetter'—those behaviors are described but not implemented in the provided scripts, so actual install activity depends on what the agent is later told to run.
Credentials
Declared requirements list no env vars or credentials, but the playbook and SKILL.md expect the operator may provide channel tokens/accountId for auto-binding. The scripts accept an --account-id and will add bindings to openclaw.json if provided, enabling a path to bind bot accounts without an explicit interactive confirmation in code. Also, materialize_team.mjs sets each agent's tools to profile:'full' and injects model/fallback provider lists; that grants broader capabilities than 'least privilege' guidance in permission-profiles and may be disproportionate to a role-generation task. The skill also describes automatic installation of other skills (which would require network/credential access) — such actions would require additional secrets but none are declared.
Persistence & Privilege
The skill is not always: true and does not request special platform privileges via metadata, but the code modifies system config (/root/.openclaw/openclaw.json), creates workspaces under /root, and writes numerous role files, including making a backup copy of openclaw.json. That is expected for a materialization tool, but is powerful: running create_team.mjs will perform persistent changes to the host environment. The code uses child_process.spawnSync to orchestrate sub-scripts, which is normal for this kind of tool but increases the surface of what a skill can do when executed.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install multi-agent-builder - 安装完成后，直接呼叫该 Skill 的名称或使用
/multi-agent-builder触发 - 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v1.0.0
Initial release of multi-agent-builder for OpenClaw.
- Enables step-by-step creation of reusable multi-agent teams based on a user’s goal.
- Automates role analysis, role confirmation, agent contract definition, and collaboration protocol generation.
- Provides an implementation-ready plan including role roster, responsibilities, provisioning requirements, and channel-binding checklist.
- Enforces explicit, dependency-aware workflows with robust safety, recovery, and security guardrails.
- Includes built-in language mirroring for user interactions (English/Chinese/others).
元数据
常见问题
Multi Agent Builder 是什么？
Build a reusable multi-agent team in OpenClaw from a user goal (e.g., "create a product-engineering team", "build a marketing ops team"). Use when the user w... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 413 次。
如何安装 Multi Agent Builder？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install multi-agent-builder」即可一键安装，无需额外配置。
Multi Agent Builder 是免费的吗？
是的，Multi Agent Builder 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Multi Agent Builder 支持哪些平台？
Multi Agent Builder 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 Multi Agent Builder？
由 gzgogo（@gzgogo）开发并维护，当前版本 v1.0.0。
推荐 Skills