Computer Use Agents
技能 活跃Build AI agents that interact with computers like humans do - viewing screens, moving cursors, clicking buttons, and typing text. Covers Anthropic's Computer Use, OpenAI's Operator/CUA, and open-source alternatives. Critical focus on sandboxing, security, and handling the unique challenges of vision-based control. Use when: computer use, desktop automation agent, screen control AI, vision-based agent, GUI automation.
To enable developers to build sophisticated and secure AI agents that can automate tasks and interact with computer interfaces by leveraging vision, reasoning, and action execution.
功能
- Perception-Reasoning-Action loop architecture
- Sandboxed environment patterns for secure execution
- Official Anthropic Computer Use implementation details
- Code examples for screen capture, mouse/keyboard control, and bash commands
使用场景
- Building AI agents for computer use from scratch
- Integrating vision models with desktop control
- Understanding and analyzing agent behavior patterns
- Deploying and testing agent behavior safely in isolated environments
非目标
- Providing a fully managed AI agent runtime
- Directly controlling user systems without sandboxing
- Handling complex enterprise-specific automation workflows without customization
Documentation
- info:Configuration & parameter referenceWhile the `SKILL.md` provides code examples and explains patterns, specific configuration parameters for the `ComputerUseAgent` class or Anthropic tools are not exhaustively documented with defaults.
Scope
- info:Minimal I/O surfaceThe code examples show structured inputs for actions and tool calls, but the exact schema for all parameters and responses isn't fully detailed in the markdown for completeness.
- info:Tool surface sizeThe `SKILL.md` describes specific patterns and provides code examples for a few core tools (`computer`, `bash`, `text_editor`), rather than exposing a large number of distinct tools.
Trust
- warning:Issues AttentionIn the last 90 days, 17 issues were opened and 4 were closed, indicating a low closure rate and potentially slow maintainer response.
Execution
- info:ValidationThe code examples show structured input for tool parameters but do not explicitly detail the use of schema validation libraries like Zod or Pydantic within the SKILL.md context.
Code Execution
- info:Error HandlingThe provided code snippets show basic error handling, but the SKILL.md does not comprehensively detail error categorization, structured reporting, or specific recovery steps for all potential failure modes.
Errors
- info:Actionable error messagesWhile the documentation emphasizes robustness, the SKILL.md does not explicitly detail the format or content of all potential user-facing error messages or provide specific remediation links.
Protocol
- info:Idempotent retry & timeoutsThe provided code examples and patterns do not explicitly detail idempotency for mutations or hard per-call timeouts on agent actions, though the Anthropic implementation might handle some of this.
Practical Utility
- info:Edge casesThe documentation highlights critical aspects like sandboxing and vision agent limitations but doesn't exhaustively list failure modes and recovery steps for every scenario.
Safety
- info:Halt on unexpected stateWhile the emphasis on sandboxing implies controlled execution, the `SKILL.md` does not explicitly detail machine-readable pre-conditions or instructions to halt on unexpected user-side states.
安装
npx skills add davila7/claude-code-templates通过 npx 运行 Vercel skills CLI(skills.sh)— 需要本地安装 Node.js,以及至少一个兼容 skills 的智能体(Claude Code、Cursor、Codex 等)。前提是仓库遵循 agentskills.io 格式。
质量评分
类似扩展
Azure Container Registry SDK for Python
100Azure Container Registry SDK for Python. Use for managing container images, artifacts, and repositories. Triggers: "azure-containerregistry", "ContainerRegistryClient", "container images", "docker registry", "ACR".
Lean Ctx
100AI 代理的上下文运行时 — 包含 59 个 MCP 工具、10 种读取模式、95+ 种 shell 模式、支持 18 种语言的 tree-sitter AST。将 LLM 上下文压缩高达 99%。用于读取文件、运行 shell 命令、搜索代码或探索目录。如果不存在，则自动安装。
Manus
100Delegate complex, long-running tasks to Manus AI agent for autonomous execution. Use when user says 'use manus', 'delegate to manus', 'send to manus', 'have manus do', 'ask manus', 'check manus sessions', or when tasks require deep web research, market analysis, product comparisons, stock analysis, competitive research, document generation, data analysis, or multi-step workflows that benefit from autonomous agent execution with parallel processing.
风险管理
100交易代理的风险管理领域知识 — 情感状态监控、仓位调整、回撤管理、交易心态检测和行为控制。在交易前检查风险、管理回撤、检测行为漂移或强制执行纪律时使用。触发词包括“风险”、“回撤”、“心态”、“仓位大小”、“手数”、“信心”、“报复性交易”、“过度交易”、“纪律”。
Setup Gbrain
100Set up gbrain for this coding agent: install the CLI, initialize a local PGLite or Supabase brain, register MCP, capture per-remote trust policy. One command from zero to "gbrain is running, and this agent can call it." Use when: "setup gbrain", "connect gbrain", "start gbrain", "install gbrain", "configure gbrain for this machine". (gstack)
Skill Creator
100Guide for creating effective skills