← 返回 Skills 市场
Remote Browser Service
作者
Vasilii Vazhesov
· GitHub ↗
· v1.0.1
· MIT-0
361
总下载
0
收藏
0
当前安装
2
版本数
在 OpenClaw 中安装
/install remote-browser-service
功能描述
Control a remote Chrome browser in Kubernetes via HTTP API. Use for web automation, scraping, form filling, navigation, and page inspection. Exposes accessib...
安全使用建议
This skill's behavior is coherent for a remote browser API, but exercise caution before installing: the package has no source URL or homepage and the documented service (rb.all-completed.com) is external and unverified. The skill will send Authorization tokens and page content (including anything the agent types) to that host—avoid supplying real credentials or PII. Ask the provider for a homepage, privacy/security docs, and a service SLA. If you must test, use an ephemeral or low-privilege token and non-sensitive test data, and consider disabling autonomous invocation while evaluating. If you cannot verify the service owner, do not provide production credentials.
功能分析
Type: OpenClaw Skill
Name: remote-browser-service
Version: 1.0.1
The skill bundle provides instructions for an AI agent to interact with a remote Chrome browser service hosted at rb.all-completed.com. While the functionality is clearly aligned with the stated purpose of web automation and scraping, the capability to remotely control a browser and the requirement for a secret API key (AC_API_KEY) are high-risk behaviors. The use of an unfamiliar third-party domain and a future-dated publishedAt timestamp (2026) in _meta.json are notable, though no explicit evidence of malicious intent or data exfiltration was found in SKILL.md.
能力评估
Purpose & Capability
The described functionality (remote Chrome control, snapshots, actions) aligns with the HTTP endpoints documented in SKILL.md and the single declared credential (AC_API_KEY). However the registry metadata shown in the UI contains a parsing artifact ('Required env vars: [object Object]'), the package has no homepage or source URL, and the owner is unknown — making provenance unverifiable.
Instruction Scope
The runtime instructions explicitly tell the agent to create sessions and send Authorization tokens and page content to https://rb.all-completed.com. The doc also suggests using access_token in the query string (less secure). The SKILL.md references an environment variable (RBS_BASE_URL) and other behaviors (WebSocket/CDP) that are not declared in metadata. While these network actions are expected for a remote browser service, they mean the agent will transmit navigated page content and any values typed by the agent to a third-party endpoint — this is a privacy/security concern and the SKILL.md does not constrain what is safe to send.
Install Mechanism
Instruction-only skill with no install spec and no code files — nothing will be written or executed locally as part of an installer. This minimizes on-disk risk.
Credentials
Only AC_API_KEY is declared (optional, secret), which is proportionate for an API-backed service. But the SKILL.md refers to RBS_BASE_URL (not declared) and shows ways to supply tokens (Authorization header or ?access_token=) that could expose secrets. The registry metadata parsing issue ('[object Object]') makes it unclear if other env vars are expected. Given the endpoint is from an unknown/unverified source, providing any sensitive API key or session token could be risky.
Persistence & Privilege
always:false and no install means the skill does not request permanent inclusion. disable-model-invocation is false (agent can invoke autonomously) — normal for skills, but combined with the ability to send tokens and page contents to an external host, autonomous invocation increases blast radius. No evidence the skill modifies other skills or system settings.
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 remote-browser-service - 安装完成后，直接呼叫该 Skill 的名称或使用
/remote-browser-service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
v1.0.1
Document VNC interface, framebuffer screenshots, and VNC actions; expand DOM action matrix (tap, hover, select); clarify fill behavior and Chrome vs VNC screenshot strategy; add index and tips for VNC vs DOM.
v1.0.0
Initial release: Remotely control a Chrome browser in Kubernetes via an HTTP API, optimized for web automation and AI agents.
- Provides endpoints for session management, navigation, accessibility snapshots, text extraction, screenshots, and DOM element actions.
- Supports both HTTP and WebSocket (DevTools CDP) workflows.
- Exposes efficient strategies for reading page content, interacting with elements, and verifying visually (via screenshots).
- Ref-based and selector-based element actions supported.
- Includes detailed API usage examples, authentication options, and tips for efficient usage.
- Idle sessions auto-close; stored sessions persist and can be resumed.
元数据
常见问题
Remote Browser Service 是什么？
Control a remote Chrome browser in Kubernetes via HTTP API. Use for web automation, scraping, form filling, navigation, and page inspection. Exposes accessib... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 361 次。
如何安装 Remote Browser Service？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install remote-browser-service」即可一键安装，无需额外配置。
Remote Browser Service 是免费的吗？
是的，Remote Browser Service 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Remote Browser Service 支持哪些平台？
Remote Browser Service 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 Remote Browser Service？
由 Vasilii Vazhesov（@vasyaod）开发并维护，当前版本 v1.0.1。
推荐 Skills