Claude 工程实践
2525 篇文章
重新设计 Anthropic 的源代码管理(source control)
Redesigning source control at Anthropic
我们如何跨产品约束 Claude
How we contain Claude across products
关于近期 Claude Code 质量报告的更新
An update on recent Claude Code quality reports
扩展 Managed Agents:将大脑与双手解耦
Scaling Managed Agents: Decoupling the brain from the hands
我们如何构建 Claude Code auto mode:一种更安全地跳过权限的方式
How we built Claude Code auto mode: a safer way to skip permissions
面向长时间运行的应用开发的框架设计
Harness design for long-running application development
Claude Opus 4.6 在 BrowseComp 表现中的评测意识(eval awareness)
Eval awareness in Claude Opus 4.6’s BrowseComp performance
用一组并行 Claude 构建 C 编译器
Building a C compiler with a team of parallel Claudes
量化智能体编程评测(eval)中的基础设施(infrastructure)噪声
Quantifying infrastructure noise in agentic coding evals
设计抗 AI 的技术评测
Designing AI-resistant technical evaluations
揭开 AI 智能体评测的面纱
Demystifying evals for AI agents
面向长时间运行智能体(agent)的有效框架(harness)
Effective harnesses for long-running agents
在 Claude 开发者平台上推出高级工具调用
Introducing advanced tool use on the Claude Developer Platform
使用 MCP 进行代码执行:构建更高效的智能体
Code execution with MCP: Building more efficient agents
超越权限提示:让 Claude Code 更安全、更自主
Beyond permission prompts: making Claude Code more secure and autonomous
用 Agent Skills 为现实世界装备智能体
Equipping agents for the real world with Agent Skills
为 AI 智能体(agent)进行有效的上下文工程(context engineering)
Effective context engineering for AI agents
对最近三个问题的复盘(postmortem)
A postmortem of three recent issues
为智能体编写有效的工具——借助智能体
Writing effective tools for agents — with agents
Desktop Extensions:为 Claude Desktop 提供一键安装 MCP 服务器
Desktop Extensions: One-click MCP server installation for Claude Desktop
我们如何构建我们的多智能体研究系统
How we built our multi-agent research system
“think”工具:让 Claude 在复杂工具调用场景中停下来思考
The "think" tool: Enabling Claude to stop and think in complex tool use situations
借助 Claude 3.5 Sonnet 在 SWE-bench Verified 上抬高标杆
Raising the bar on SWE-bench Verified with Claude 3.5 Sonnet
构建高效的智能体
Building effective agents
引入上下文检索(Contextual Retrieval)
Introducing Contextual Retrieval