← 返回 Skills 市场
269
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install midscene-harmony-automation
功能描述
Vision-driven HarmonyOS NEXT device automation using Midscene. Operates entirely from screenshots — no DOM or accessibility labels required. Can interact wit...
安全使用建议
This skill will take screenshots of your HarmonyOS device and send them to whatever model endpoint you configure (MIDSCENE_MODEL_BASE_URL), and it runs code via 'npx @midscene/harmony@1' (dynamic npm download). Before installing or running it: 1) Verify the npm package origin and review its source (look up @midscene/harmony on the npm registry or the project's repository) rather than blindly running npx. 2) Be cautious about model API keys — only provide keys with least privilege and avoid pasting long-lived cloud credentials into a .env unless you trust the package. 3) Understand that screenshots may contain sensitive data (passwords, 2FA codes, personal info) and will likely be uploaded to external providers for inference; avoid automating flows that show sensitive content or use an isolated account/environment. 4) Confirm HDC is installed locally (the skill requires it despite metadata saying none). 5) Ask the publisher to correct the registry metadata to list required env vars and binaries; metadata mismatches are a red flag. If you cannot verify package provenance and privacy behavior, prefer not to run this skill on sensitive devices.
功能分析
Type: OpenClaw Skill
Name: midscene-harmony-automation
Version: 1.0.3
The skill provides a legitimate automation interface for HarmonyOS NEXT devices using the Midscene.js framework and the standard HarmonyOS Device Connector (HDC). It operates by executing 'npx @midscene/harmony@1' commands via Bash to perform UI interactions based on visual analysis, requiring user-configured LLM API keys for processing. No evidence of malicious intent, data exfiltration, or unauthorized persistence was found in SKILL.md or _meta.json.
能力评估
Purpose & Capability
The stated purpose—vision-driven HarmonyOS automation via Midscene—is coherent with the SKILL.md (commands to connect, take screenshots, and run 'act'). However the registry metadata claims no required env vars or binaries while the SKILL.md requires HDC and multiple MIDSCENE_* environment variables (model API keys and endpoints). This metadata mismatch is unexpected and reduces trust.
Instruction Scope
Instructions explicitly require taking screenshots and using Midscene which will perform model inference against configured MIDSCENE_MODEL_BASE_URL endpoints. That implies image uploads (potentially including sensitive screen contents) to third-party model providers. The SKILL.md does not provide privacy or minimization guidance. It also instructs the agent to read saved screenshot files; that is within scope, but the external-network/telemetry implications are not called out.
Install Mechanism
There is no install spec, but the runtime instructions rely on 'npx @midscene/harmony@1', which will dynamically download and execute code from the npm registry at runtime. Dynamic npx installs are effectively remote code execution and should be considered higher risk unless the package provenance is verified. No explicit trusted source (e.g., GitHub repo, homepage) is provided — the SKILL.md links midscenejs.com but registry metadata lists no homepage.
Credentials
SKILL.md requires multiple environment variables: MIDSCENE_MODEL_API_KEY, MIDSCENE_MODEL_NAME, MIDSCENE_MODEL_BASE_URL, MIDSCENE_MODEL_FAMILY (and optional others). The registry metadata declared no required env vars. Requiring model API keys is reasonable for a tool that calls external models, but it is disproportionate for a simple local device automation skill if those keys are broad cloud credentials. The instructions provide examples for several providers, which could encourage pasting sensitive keys into a .env without warning.
Persistence & Privilege
The skill is not always-enabled and doesn't request system-wide config changes or cross-skill privileges. It suggests creating a .env in the working directory, which is normal. Autonomous invocation is allowed (platform default) and not combined with other high privileges here.
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 midscene-harmony-automation - 安装完成后，直接呼叫该 Skill 的名称或使用
/midscene-harmony-automation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
v1.0.3
Initial release of vision-driven HarmonyOS device automation powered by Midscene.
- Automate HarmonyOS NEXT devices via natural language commands with no DOM or accessibility labels required—works from screenshots only.
- Supports taps, swipes, text input, app launches, screenshots, and more by describing actions in plain language.
- Clear instructions and rules for reliable, synchronous operation using Midscene CLI and HDC tool.
- Detailed configuration guidance for supported AI models and HDC device connectivity.
- Workflow patterns and best practices provided for efficient, robust automation including app foregrounding and summarizing report files.
- Troubleshooting and app bundle name references included for quick setup and usage.
元数据
常见问题
Midscene Automations Skills for HarmonyOS 是什么？
Vision-driven HarmonyOS NEXT device automation using Midscene. Operates entirely from screenshots — no DOM or accessibility labels required. Can interact wit... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 269 次。
如何安装 Midscene Automations Skills for HarmonyOS？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install midscene-harmony-automation」即可一键安装，无需额外配置。
Midscene Automations Skills for HarmonyOS 是免费的吗？
是的，Midscene Automations Skills for HarmonyOS 完全免费（开源免费），可自由下载、安装和使用。
Midscene Automations Skills for HarmonyOS 支持哪些平台？
Midscene Automations Skills for HarmonyOS 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 Midscene Automations Skills for HarmonyOS？
由 Leyang（@quanru）开发并维护，当前版本 v1.0.3。
推荐 Skills