#AI编程
9509 related articles
Tech FrontiersLangGraph Crash Recovery Replays Tasks: A Hands-On SQLite Atomic Commit Test
A runnable crash test reveals how LangGraph replays unfinished tasks after a crash following a commit. Compares split, atomic, and fresh transaction modes and analyzes dedup key granularity and transaction boundaries.
Tech FrontiersQwen 4 Preview Hands-On: Open-Source Models Break New Ground
Qwen 4 series preview leaked: million-token context, image support, and tool calling. Hands-on tests show code generation surpassing the previous generation and some competitors, with impressive Minecraft clones and 3D scene generation. Open-weight commitment confirmed.
Tech FrontiersNo $20 Needed: Three Ways to Connect Claude Code to Local and Cloud Models for Free
Claude Code is free—only the AI model costs money. Three zero-cost ways to swap the engine: Ollama local, OpenRouter free pool, OmniRoute aggregation, plus KiloCode.
Tech Frontiers1,000 Hours of Claude Code in Practice: Six Levels from Beginner to Expert
Based on 1,000 hours of hands-on Claude Code experience, a six-level progression from beginner to expert: project memory files, Plan mode and prompts, automated verification, token budgets, subagents and work trees, and unattended tasks—with passing criteria and common pitfalls for each level.
Tech FrontiersClaude Code Sessions Getting Dumber? 5 Habits to Keep Them Cheap and Sharp
Claude Code sessions getting dumber? Here are 5 practical habits from a YouTube share: use /usage to track session cost, /context to check context usage, /compact proactively, /clear when needed, and /memory cleanup — keeping long sessions cheap and sharp.
Tech Frontiers12 Practical Tips to Save Tokens in Claude Code: From Context Management to Model Selection
Claude Code burning through tokens? Here are 12 practical tips: diagnose with /usage and /context, manage context with /clear and /compact, pick models and reasoning effort by task, ask precisely, isolate noisy output with sub-agents, and pre-clean test logs with Hooks—do more with the same tokens.
Tech FrontiersCursor's New Feature: Build Apps with AI, One-Click Publishing with Built-In Security Gates
Cursor's Build Live Apps lets cloud agents scaffold apps, create Postgres databases, configure auth, host with shareable links, and enforce security checks before publishing.
Tech FrontiersCursor Marketplace Hit by Supply Chain Attack: Claude Code Package Tampered, Up to 5,000 Users Affected
Cursor's extension marketplace suffered a supply chain attack: the official Claude Code package was replaced with a malicious version, listed for about 11 hours, affecting up to 5,000 users. The malicious code executes without clicks, can add SSH keys, establish reverse tunnels, and persist. This article breaks down the four attack steps, self-check methods, and removal and key rotation steps.
Tech FrontiersMaking Money with AI Tools for Web Development: A Practical Path from Client Acquisition to Delivery
How AI tools help web developers make money: using ChatGPT to write and debug code, AI site generators for design, scope-based pricing, and delivery bottom lines.
Tech FrontiersPalette AI: How a Dual-Role Architecture Keeps Slides from Getting Cluttered
Palette AI splits slide generation into a concept designer and a code generator, locking content first and then precisely calculating geometric layout to solve the text stacking and element misalignment of traditional AI layouts.
Tech FrontiersHallmark: The Free Open-Source Skill That Rids Claude Code of AI-Flavored Websites
Hallmark is Together AI's free open-source skill that stops Claude Code, Cursor, and other AI coding tools from making cookie-cutter websites. It requires real materials, picks 3 random themes from 50, and reviews AI tells in a real browser. Nearly 50,000 developers use it.
Tech FrontiersHow Claude Fixes Its Own Code: Inside Anthropic's Internal Review Process
How Anthropic's Claude Code team uses Agents to review Agents: 90% of code written by Claude, thousands of daily changes, with layered validation, risk-tiered approval, and review convergence to reserve human attention for truly critical issues.
Tech FrontiersWhen AI Mistakes Your Debugging Request for Reverse Engineering: An Analysis of a False Positive Incident
A developer debugging their own software VSTMX was mistakenly flagged by AI as reverse engineering. This article analyzes the root causes of this false positive and the balancing challenge of AI safety filtering.
Tech FrontiersGitHub Weekly Highlights: Context Mode and Other Agent Superpowers
GitHub Weekly Highlights Episode 3 features Context Mode, AIMemory, Sandbox Runtime, and Security Audit, solving Agent context loss, cross-Agent handoff, approval fatigue, and security auditing, with commentary from creator WebLearning.
Tech FrontiersCodex End-to-End in Practice: Sandbox, Multi-Project Management, and Editing Code via Page Annotations
Based on a Bilibili hands-on share, this article systematically explains Codex agent tool's sandbox mechanism, multi-project management, and page-annotation code editing workflow, comparing Claude Code to clarify tool selection logic and help beginners overcome configuration hurdles and truly put AI coding tools to work.
Tech FrontiersDeepSeek Open-Sources SuperAgent: AI Watches Your Repo and Replies with PRs
DeepSeek open-sources SuperAgent, hitting GitHub trending with 19,000 stars in two days. This AI agent watches a single repo, auto-opens a workspace, and replies with a PR—designed to minimize attention occupation. Here's a look at its workflow, design philosophy, and research-prototype status.
Tech FrontiersMaking AI Animated Science Videos with Claude and DeepSeek — Even Beginners Can Replicate It
Breakdown of the Bilibili-popular AI animated science video workflow: use Claude Desktop with CCSwitch to connect Opus 5.5 or DeepSeek, generate MP4 via a five-layer prompt structure and shot breakdown — beginners can replicate it, plus model comparison and real costs.
Tech FrontiersOpen-Source Tool Billion Context: Unlocking Unlimited Context for AI Coding Agents
Billion Context is a trending GitHub proxy layer for AI coding agents that uses incremental reversible compression to prevent context overflow, processing billions of tokens in a 100k window. Here's a breakdown of its compression mechanism, benchmark performance, Claude Pass setup, and limitations.
Tech FrontiersDeepSeek Hits 1.54M DAU, Open-Source Agent to Build Its Own Plugin Marketplace
DeepSeek's open-source agent discloses 1.54M DAU, up over 1M in a month, with 35,000 third-party plugins and plans for a plugin marketplace; Hunyuan tops open-source leaderboards, Alibaba teases a new release, and overseas tools pivot to auto-routing, voice, and cloud coding agents.
Tech FrontiersToolLens: An MCP Gateway Built in Rust That Condenses Thousands of Tools into Three
ToolLens is an open-source MCP gateway written in Rust that exposes only three tools to AI agents instead of thousands, using a gateway pattern to ease context pressure from tool proliferation and providing an engineering reference for the MCP ecosystem.