by HELLO EVERYONE!!! It’s March 20th, 2026, and you are reading the 103rd edition of Codeminer42’s tech news report. Let’s check out what the tech world showed us this week!
The Miners’ post of the week 🧑🏻💻👩🏽💻
AI TIP: Running Two Claude Code Accounts — by The Miners
This post shares a practical tip for running multiple Claude Code accounts simultaneously using the CLAUDE_CONFIG_DIR environment variable. By creating separate config directories and using bash aliases, you can maintain completely isolated accounts for personal and work contexts without constantly logging out. The solution leverages how Claude Code respects environment variables for configuration paths, enabling seamless context-switching in multiple terminals. This technique extends to unlimited accounts and applies to any shell environment.
NVIDIA DLSS 5 Delivers AI-Powered Breakthrough in Visual Fidelity for Games
This interesting article introduces NVIDIA DLSS 5, the company’s most significant graphics breakthrough since real-time ray tracing in 2018. DLSS 5 uses AI to infuse pixels with photoreal lighting and materials, bridging the gap between real-time rendering and cinematic quality. Arriving this fall, it’s supported by major publishers including Bethesda, CAPCOM, and Ubisoft, enabling game developers to achieve Hollywood-level visual effects in real-time interactive gameplay.
Leanstral: Open-Source foundation for trustworthy vibe-coding
This article announces Leanstral, the first open-source code agent for Lean 4, designed to formally prove code implementations against strict specifications. With just 6B active parameters, Leanstral outperforms much larger open-source models like GLM-5 while costing a fraction of closed-source alternatives like Claude Opus. Released under Apache 2.0, it’s available through Mistral Vibe, a free API endpoint, and as downloadable weights, addressing the critical bottleneck of human code review in high-stakes domains.
Petition to Node.js TSC: No AI code in Node.js Core — by Fedor Indutny
This interesting post documents a petition to the Node.js Technical Steering Committee opposing LLM-assisted pull requests in core. Sparked by a 19k-line PR disclosed as Claude Code-generated, the petition questions whether AI-generated contributions satisfy the Developer’s Certificate of Origin and risks diluting Node.js’s reputation for hand-crafted, carefully reviewed code. With 39+ signatories, including TSC members and core contributors, it’s reigniting important conversations about code authenticity and quality in critical infrastructure.
LLM Architecture Gallery — by Sebastian Raschka, PhD
This interesting article presents a comprehensive gallery of LLM architecture figures and fact sheets spanning from GPT-2 to cutting-edge 2025 models like Kimi K2 (1T) and Grok 2.5. Created by Sebastian Raschka, it collects detailed architecture panels showing how model design has evolved, including dense vs. sparse MoE approaches, attention mechanisms, and parameter counts. Available as downloadable figures and physical posters, it’s an invaluable reference for understanding the architectural patterns that define modern large language models.
A couple of git nits — by Estib Vega
This article is a thoughtful rant on Git’s UX quirks, exploring why commonly-used combinations like git log --oneline --graph --decorate --all aren’t defaults, why git cat-file -p requires knowing the object type first, and the discovery problem with obscure flags like -S vs -G. The author argues that when an entire ecosystem independently converges on the same workaround, it signals design failure. While acknowledging Git’s critical role and historical context, they advocate for complaining and iterating toward better tooling, citing Git itself as proof that dissatisfaction drives progress.
The rise of malicious repositories on GitHub — by Artem Golubin
This article documents a surge of malicious repositories on GitHub that mimic legitimate projects but distribute Windows malware. The author discovered over 100 such repos using a simple GitHub search pattern, some LLM-generated to boost search rankings, others with hijacked accounts. Notably, some fake MacOS/Linux projects only provide Windows binaries, suggesting automated campaigns. Despite reports to GitHub, these repositories persist for months with constantly-updated READMEs. Browsers block most downloads, but the article emphasizes always verifying repository legitimacy before downloading.
OpenAI JavaScript/TypeScript Library
This article documents the official OpenAI API library for JavaScript and TypeScript, generated from OpenAI’s OpenAPI specification. It showcases the new Responses API as the primary interface for text generation, while maintaining support for the legacy Chat Completions API indefinitely. The library supports streaming, file uploads, webhook verification, and the Realtime API for low-latency multi-modal conversations over WebSocket. Available via npm, JSR, and Deno, it provides comprehensive error handling with specific exception types for different HTTP status codes.
jsdom: JavaScript implementation of web standards
This post introduces jsdom, a pure-JavaScript implementation of WHATWG DOM and HTML standards for Node.js testing and web scraping. It provides a comprehensive sandbox for parsing HTML, executing JavaScript with optional script execution, and simulating browser APIs like localStorage and sessionStorage. The library supports customization through options like URL context, content type, and visual browser emulation, while emphasizing security considerations when running untrusted code with runScripts settings.
Node.js worker threads are problematic, but they work great for us
This post explores Node.js worker threads through Inngest’s experience solving event loop starvation in their WebSocket-based Inngest Connect. Worker threads provide isolation between CPU-bound user code and critical infrastructure like heartbeats, but come with constraints: they can’t receive function closures (only file paths), communicate via costly message serialization, and carry ~10MB memory overhead each. The article compares Node’s threading model to Go goroutines and Rust threads, discussing why these constraints exist and how Inngest worked around them in their TypeScript SDK.
Building Production-Ready Multi-Tenant SaaS in Rust with Actix-Web
This article explores building scalable multi-tenant SaaS applications using Rust and the Actix-Web framework. It covers architectural patterns, isolation strategies, and performance considerations specific to running multiple customer workloads safely within a single application. The guide addresses database design for multi-tenancy, request routing, authentication, and resource limits, providing practical patterns for teams building production SaaS on Rust’s performance and safety guarantees.
Why Node.js Needs a Virtual File System
This post examines the case for virtual file systems in Node.js, exploring how filesystem abstractions could improve testing, sandboxing, and development workflows. The article discusses the challenges of current file I/O patterns and how a VFS could enable better isolation between applications, safer execution of untrusted code, and more flexible development environments that don’t depend on the host operating system’s filesystem.
React SSR Framework Benchmark: TanStack Start, React Router, Next.js
This post benchmarks modern React SSR frameworks, including TanStack Start, React Router, and Next.js, comparing performance metrics, developer experience, and architectural decisions. The analysis evaluates startup times, build performance, rendering efficiency, and runtime characteristics across different use cases, helping developers choose the right framework for their specific needs.
Languages, Tools & Framework releases
Next.js 16.2
This post announces Next.js 16.2 with significant performance improvements, including ~400% faster dev startup time and ~50% faster rendering. The release features improved debugging with hydration diff indicators, Server Function logging, and over 200 Turbopack fixes. New capabilities include experimental error recovery with unstable_catchError() and experimental retry logic in error boundaries, alongside stable Adapters for platform customization and better support for multiple icon formats.
Bun v1.3.11 — by Jarred Sumner
This release post details Bun v1.3.11, which shrinks the Linux binary by 4MB and introduces Bun.cron for OS-level cross-platform scheduling compatible with Cloudflare Workers. New utilities include Bun.sliceAnsi for terminal-aware string slicing and enhanced markdown rendering with richer metadata. The release fixes critical UDP socket bugs on macOS and improves fs.stat sub-millisecond precision, while adding –path-ignore-patterns for smarter test discovery and native ARM64 support for Windows node_modules/.bin binaries.
Herb v0.9 — by Marco Roth
This post announces Herb v0.9, the modern HTML+ERB toolchain, with a major focus on stability and practical features. The headline feature is Action View tag helper support, allowing the parser to understand tag.*, content_tag, and link_to calls as equivalent HTML elements. New built-in rewriters enable conversion between Action View helpers and plain HTML, while the language server gains hover documentation and code actions. The release includes full arena allocator integration and fixes numerous parser bugs, with record community involvement across 198 commits from 13 contributors.
Ruby 4.0.2
This article announces Ruby 4.0.2, a routine update that includes a bugfix in YJIT for NoMethodError on Puma. Following a two-month release cadence, Ruby 4.0.3 is planned for May with subsequent releases in July, September, and November. The release maintains stability with carefully vetted patches, continuing the momentum of Ruby 4.0’s introduction of Ruby Box and ZJIT, building on the language’s commitment to performance and reliability improvements.
—
And that’s all for this week! Wish you all a great weekend and happy coding!
We want to work with you. Check out our Services page!
