#AI编程
9355 related articles
Malicious Claude Code Extension Infiltrates Cursor: Version 2.1.313 Self-Check and Cleanup Guide
A malicious Claude Code build 2.1.313 appeared in the Cursor extension marketplace for about 11 hours, affecting up to 5,000 users. StepSecurity disclosed that it stole system information, git email, and environment variable names, and attempted to implant SSH keys and reverse tunnels. This article covers the incident, malicious behavior, and complete self-check and cleanup steps.
15 Essential Prompts for Vibe Coding: Keep AI from Wrecking Your Code
Vibe Coding often tempts AI to break your code. These 15 prompts—from PRD, architecture, and database to testing and debugging—chain each output as the source of truth, turning AI coding from casual chat into a controllable engineering process.
Cursor vs Windsurf: Code Review or Direct Apply — Which Fits You Better?
Cursor shows a diff for review before changing code; Windsurf applies changes directly. We analyze how each mode affects your workflow and cover Cursor 2.0's multi-agent mode.
Free AI API Keys in Practice: A Zero-Cost Guide to Calling GPT, Claude, Gemini, and DeepSeek
Get free API keys for GPT, Claude, Gemini, and DeepSeek without a credit card. This guide covers TokenRouter, Semantic Nova, Hugging Face, Grok, and Token Harbor integrations with VS Code and coding agents, plus a reality check on free quota limits.
Claude Extension Backdoored for Nearly 12 Hours, Cursor Users Should Be Alert
The Claude-related extension on the Cursor platform was implanted with a backdoor for nearly 12 hours, affecting up to about 5,000 developers. The malicious code attempted to install an SSH backdoor and steal cloud credentials, and an official fix has been released. Learn the details and response recommendations.
Add Scripts to Cursor Skills in 45 Seconds: A Three-Step Tutorial
Cursor Skill tutorial: in three steps, add a scripts folder to skill.md, point to scripts with relative paths, and use references and assets to manage docs and templates so the Agent actually runs Python and Bash scripts rather than just describing them.
Deleting the Extension Isn't Enough? A Post-Mortem of the Poisoned Claude Code Extension Incident
The Claude Code extension was poisoned in Cursor, with two builds online for ~11 hours, potentially reaching 5,000 developers. Anomalous windows and an implanted SSH key persisted after deletion. Cursor warns uninstalling may not be enough—wipe the machine or run five checks.
From Programmer to AI Architect: The Silent Shift in the Programming Role
AI coding tools like Cursor are transforming software development: developers shift from manual syntax to high-level system architecture. This article analyzes the efficiency gains, technical debt risks, and the trend of architectural judgment becoming the new scarce resource.
ClaudeMD: The Config File That Makes AI Code Assistants Follow Your Rules
ClaudeMD is a project-root config file that lets AI coding assistants like ClaudeCode and Cursor auto-read team conventions—naming rules, test commands, forbidden patterns—at session start, enabling versioned, shareable rules.
One Command Does It All: Connect Relia to Cursor, VS Code, and Other Mainstream AI Coding Tools
The official Relia tutorial shows how one command connects Relia to Cursor, VS Code, Windsurf, Codex, and Gemini CLI, auto-detecting installed tools, writing MCP configs, and supporting project-level setup and connection management.
Agent Skills Standard Transferred to Linux Foundation, Plus Three Updates to AI Coding Tools
Anthropic transfers the Agent Skills standard to the Linux Foundation's Agentic AI Foundation with nearly 280 million installs; Cursor releases an Agent SDK for agents-as-code; Cairo CLI 2.33 adds folder trust gating; Devin Desktop disables the agent diff staging area by default. Four news items outline the standards, tooling, security, and interaction layers of agentic coding infrastructure.
Superpowers: An Open-Source Framework That Teaches AI Coding Agents to Think Like Engineers
Superpowers is an open-source methodology for AI coding agents that enforces brainstorming, planning, TDD, and code review through composable skills—supporting 17 platforms including Claude Code, Cursor, and Devin.
Connecting Docsie to Devin via MCP: Full Configuration and Hands-On Walkthrough
Step-by-step guide to connecting Docsie to Devin via MCP: add a custom MCP server, configure authorization and write permissions, with three hands-on scenarios covering video-to-docs, PII compliance scanning, and workspace overview, plus capability and cost boundaries.
Companies Selling Tokens Are Now Scoring Your Coding Efficiency
AI coding assistants now score developer efficiency: changing one line is rated inefficient, while having AI rewrite the whole app daily tops the leaderboard. The rules are set by the company selling tokens—a bias every developer should watch.
Google Shuts Down Jules AI: The Trust Crisis of AI Coding Agents
Google will shut down AI coding agent Jules in December 2026 with no appeals or pivot. Meanwhile Devin gains enterprise governance and Cursor expands fast—but fake GitHub repos are tricking AI agents into installing malware. As infrastructure strains and platforms scramble to scale, how should developers choose a trustworthy coding agent?
What Is an AI Software Factory? The Next Stop in Development Paradigms
AI software factories are becoming the new trend after chat-based programming: AI agents handle planning, coding, testing, and review, while humans stay in the loop to describe outcomes. This article analyzes the approaches of Cognition, Factory, and NVIDIA, and the engineering prerequisites behind them.
Can Open-Source Dify Challenge Devin? A 60-Second Install Test
Can open-source Dify challenge Devin? Based on a hands-on video, we break down Dify's 60-second install from blank terminal to runnable build, its AI workflow orchestration, RAG pipelines and agent capabilities, and the value and limits of its 100% open-source, zero-subscription model versus commercial SaaS tools.
How Does the AI Coding Agent Work? Unraveling the Loop Mechanism Behind Text-Model Code Fixes
An in-depth look at how AI coding agents work: how text models use the observe-decide-act loop, tool calls, context management, and test judges to fix real software, plus their limits and permission design.
AI Engineering Hub: A Collection of Hands-On LLM and RAG Tutorials
GitHub project ai-engineering-hub focuses on in-depth tutorials for LLMs, RAG, and AI Agents, offering runnable code as Jupyter Notebooks with nearly 40,000 stars—ideal for developers building AI applications.
When AI Writes Everything: The Fundamental Skills Crisis Facing New Graduates and How to Break Through
A new graduate warns after observing AI teaching at his alma mater: over 60% of students can't use arrays or integrals yet rely on AI for all their code. This article analyzes the causes of the fundamentals gap and offers guidance from a senior engineer's perspective.