← 返回 Skills 市场
Workhorse Duo
作者
wangyangwjy
· GitHub ↗
· v2.1.2
· MIT-0
90
总下载
1
收藏
0
当前安装
5
版本数
在 OpenClaw 中安装
/install workhorse-duo
功能描述
Run a practical two-worker orchestration workflow with Xiaoma (execution) and Xiaoniu (QA). Use when the user wants 小马执行 / 小牛验收, when a medium/large task sho...
安全使用建议
This skill implements a legitimate execute->QA local workflow, but it assumes and performs privileged local actions that are not declared in the top-level metadata. Before running any bootstrap helper or using AutoFixConfig: 1) Verify you have the openclaw CLI installed and you understand the commands; 2) Inspect the PowerShell helper and only run it manually if you trust the source; 3) Prefer running the bootstrap with AutoFixConfig disabled, check the current ~/.openclaw/openclaw.json yourself, and make any changes manually so you control exactly what fields change; 4) Be cautious about accepting agentToAgent.allow = ["*"] and sessions.visibility = "all" — these widen cross-agent visibility and can increase data-sharing/exposure; tighten policies after validation; 5) Keep backups of openclaw.json (the helper makes a .bak) and know how to restore/restart the gateway if needed. The package should have declared required binaries/config paths — absence of that declaration is an omission worth questioning with the publisher.
功能分析
Type: OpenClaw Skill
Name: workhorse-duo
Version: 2.1.2
The skill bundle implements a multi-agent orchestration workflow but includes a PowerShell bootstrap script (in published-bootstrap-helper.md) that modifies the local OpenClaw configuration. Specifically, it enables broad cross-agent communication permissions (agentToAgent.allow = ['*']) and performs a service restart (openclaw gateway restart). While these actions are documented as setup requirements and accompanied by rollback instructions in risk-and-rollback.md, the automated modification of security-sensitive configurations and service management constitutes a high-risk capability.
能力评估
Purpose & Capability
SKILL.md and references clearly depend on the local openclaw CLI and on editing ~/.openclaw/openclaw.json and restarting the gateway to enable cross-agent routing. The registry metadata declares no required binaries or config paths, which is inconsistent: a Workhorse Duo deployment legitimately needs the openclaw CLI and access to the local openclaw config, so the package should have declared those requirements.
Instruction Scope
Runtime instructions include concrete operations that read and may modify local configuration, create agents, run CLI commands (openclaw agent/agents/gateway), and optionally auto-patch config and restart a local gateway. Those steps are within the skill's functional scope (bootstrapping local agents) but they grant broad runtime powers and perform persistent system mutations not visible in the top-level metadata.
Install Mechanism
This is an instruction-only skill with no install spec or remote downloads. That reduces supply-chain risk — there is no archive to fetch/execute — but the included PowerShell helper is intended to be copied and run locally, which places responsibility on the operator to inspect it.
Credentials
The skill asks (in the body) to modify ~/.openclaw/openclaw.json and to set bootstrap-safe values including agentToAgent.allow = ["*"] and sessions.visibility = "all". Those settings broaden cross-agent visibility/permissions and can increase data exposure between agents. The package declares no required credentials or config paths, so the implied need to access and change local config is not reflected in declared requirements.
Persistence & Privilege
Although always:false and the skill is not autonomously privileged by metadata, the provided bootstrap helper will write to a persistent config file (~/.openclaw/openclaw.json), create a .bak, and restart the gateway. That gives the skill the ability (when an operator runs the helper) to change runtime behavior system-wide. This is justified by the skill's purpose but is a material privilege and should be explicitly disclosed in metadata and UI prompts.
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 workhorse-duo - 安装完成后，直接呼叫该 Skill 的名称或使用
/workhorse-duo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
v2.1.2
Packaging hardening: published an inspectable markdown bootstrap helper, updated bootstrap references to point at bundled package files, and removed reliance on script-file packaging for first-run verification.
v2.1.1
Packaging fix: included the published bootstrap helper file and updated docs to reference the bundled root-level bootstrap script so the installed package matches the setup instructions.
v2.1.0
Workhorse Duo 2.1: clarified core value, added best-fit vs not-a-fit guidance, and made the skill's advantages more explicit for operators evaluating whether the workflow matches their environment.
v2.0.0
Workhorse Duo 2.0: improved bootstrap UX, added first-run path guidance, formalized proactive async completion reporting, and clarified personality-forward vs generic role mapping for Xiaoma and Xiaoniu.
v1.0.0
Initial release: real-agent dual-worker orchestration with Xiaoma execution, Xiaoniu QA, bootstrap, optional auto-fix, readiness checks, and async boss-mode routing.
元数据
常见问题
Workhorse Duo 是什么？
Run a practical two-worker orchestration workflow with Xiaoma (execution) and Xiaoniu (QA). Use when the user wants 小马执行 / 小牛验收, when a medium/large task sho... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 90 次。
如何安装 Workhorse Duo？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install workhorse-duo」即可一键安装，无需额外配置。
Workhorse Duo 是免费的吗？
是的，Workhorse Duo 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Workhorse Duo 支持哪些平台？
Workhorse Duo 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 Workhorse Duo？
由 wangyangwjy（@wangyangwjy）开发并维护，当前版本 v2.1.2。
推荐 Skills