Weekly picks · Week 44, 2026
Agent use cases
Collaborate with multiple AIs for coding and review using the Herdr desktop app
After installing the Herdr desktop app, Claude Code users can let Opus 5.5 coordinate up to 10 working threads through the app, with Claude Code writing the code and Codex Astra performing code reviews.
10 essential Claude Code skills and workflow optimization methods
This article introduces 10 must-have Claude Code skills from overseas that have earned over 1,200 upvotes and come from GitHub repositories with over 100k stars. By integrating these open-source skills into Claude Code and combining them with specific tasks and prompts, you can significantly boost AI performance.
Replicating Stripe's internal AI agent workflow
Handle repetitive tasks using the same AI agent setup used internally by Stripe, reducing task completion time from 20 days to 4 days compared to standard Claude Code. This system is built using over 100 saved prompts (one per step) with multiple agents collaborating to handle the build work.
Build a multi-AI scheduled relay tutorial writing pipeline
Use a private GitHub repository as a central bus to coordinate Muse for topic selection, Claude for hands-on drafting, and GPT for independent review. This enables automated daily relay production of tutorials, with final publishing gated by human review.
Run multiple AI agents in parallel using the open-source tool T3 Code
Using T3 Code, an open-source tool by Theo, you can run multiple agents like Claude Code, Codex, and Cursor simultaneously within a single interface, using your existing accounts without new subscriptions. It supports multi-machine sync, project organization, visual diffs, and parallel sessions.
Build multi-agent orchestration and visual workflows with Claude Code foreman plugins
Developers built two open-source plugins: foreman turns the main session into a coordinator solely responsible for dispatching and reviewing to keep the context clean and run sub-agents in parallel, while foreman-vision provides a pixel-art office in the browser to view agent statuses and terminals in real time.
Collaborate autonomously and assign tasks between AI agents using open-source app BroCrew
A developer built an open-source desktop app called BroCrew that lets AI agents like Claude and Codex delegate work to each other, maintain separate memories, and only ask humans for help when stuck, reducing the coordination burden of multitasking.
Build a local offline manga translation Android app with Claude Code
A developer built Yomu, an app that recognizes Japanese manga speech bubbles on a phone and translates them into multiple languages using a local large language model (llama.cpp). The entire process runs locally without internet access to protect user privacy.
Building Greia, a child-specific operating system with Claude
A developer combined Claude with old hardware to build Greia, a fully functional operating system for children featuring around 45 activities and a parental management interface. Through LLM agents, parents with no programming experience can use natural language to customize dedicated learning or gaming activities for their children.
Simulate game balance with Claude Code and AI bots
Developers used Claude Code and Opus 5.5 to build bot players with four distinct playstyles and an economic simulator, stress-testing difficulty curves and game balance before human player testing.
Build a reliable AI coding agent engineering environment
Learn Harness Engineering through an open-source course to solve common AI coding agent issues like broken code, lost context, and false reporting. Featuring 14 lessons and 8 hands-on projects, this guide teaches you how to build a reliable environment from scratch with support for persistent memory, test validation, and automated loops.
Anthropic internal AI coding architecture in practice
Learn how to boost AI agent coding efficiency using dynamically routed CLAUDE.md files, multi-terminal parallel sessions, and automated validation tests, based on hands-on insights from Claude Code core author Boris Cherny.
Build formally verified web apps using Claude and Lean
The author used Claude Opus 5.5 under high-level guidance to develop Lasm, bringing the formally verifiable Lean language to browsers, Node, serverless, and edge environments. This enables developers to build end-to-end web applications with formally verified specifications (such as Express or Hono servers running on Cloudflare Workers).
Incorporate engineering standards into AI coding with HelixHarness
Inject your team's business knowledge, engineering standards, and development workflows into AI coding agents by installing the HelixHarness plugin and running initialization and proposal commands in your project repository. It supports full lifecycle management from requirements, implementation, and testing to archiving, while enabling continuous accumulation and reuse of knowledge.
Clean up unused Claude Code skills to save tokens
The author solved the problem of rapid context bloat and token consumption caused by accidentally triggered features by disabling built-in Claude Code skills and plugins that are never used. This effectively streamlines the session context and prevents unnecessary resource waste.
Prevent AI browser testing from generating junk files and crashing your Mac
When AI agents run headless Chrome for UI testing, interrupted tests can leave behind massive amounts of junk files, causing Macs to fail bootup due to 41 million files in the temp directory. This issue can be prevented by setting up isolated temp directories, using cleanup traps, and optimizing Chrome flags.
Build and run iOS apps on iPhone using AI agents without a Mac
The author used a Cursor cloud agent to build, install, launch, and drive an iOS app directly on an iPhone, completely eliminating the need for a Mac computer, data cables, or simulators. While this establishes an end-to-end workflow, the current setup is still in its early stages and requires further refinement.
Build a personal intelligence agent with Claude Code
Using Claude Code along with CLAUDE.md and custom collection and analysis skills, a user built a personal intelligence agent to monitor specific topics and surface insights they might have missed. After about a week of use, this cut an hour of daily browsing down to a 2-minute focused briefing.
Build an AI agent system from a blank terminal in 10 minutes
Build an AI agent system from a blank terminal in under 10 minutes using Claude Code, scaling from a single agent to automatic execution, checking, iteration, and multi-agent collaboration. Reportedly, 95% of engineers inside Anthropic use a similar workflow.
Explore 10 Claude Code skill repositories
This article lists 10 Claude Code skill repositories to help users access a rich library of skills for extending Claude Code capabilities.
Replicating Stripe's internal AI agent workflow
Stripe shared its internal AI agent setup, which cut task turnaround times from 20 days to 4 days and reduced payment integration time from up to 6 months to 2-6 weeks. The approach achieves efficient automation by breaking repetitive work into modular prompts, using dual agents for mutual oversight, and implementing self-updating prompts.
Create digital human podcasts and e-commerce videos with AI Agents
Using an AI Agent combined with Renoise MCP and Remotion, you can fully automate the generation of digital human videos featuring scripts, lip-syncing, audio processing, subtitles, and professional-grade packaging using just a single photo and a one-sentence input. It comes with 5 built-in editing styles, 48 creator-grade packaging components, and multi-language support.
Turn a single AI session into a 49-agent game studio using an open-source project
This open-source project introduces a 49-professional-AI-agent hierarchical game studio architecture to Claude Code, featuring directors, department leads, and various specialists. Through structured collaboration, 74 workflow skills, and built-in QA and release processes, it automates everything from brainstorming to game publishing.
Fully automated 7-agent independent website outsourcing system based on Claude Code
A solo operator uses Claude Sonnet 4.6 to build a 7-agent automation system that automatically scans Google Maps for small businesses lacking websites, handling diagnosis, landing pages, videos, and personalized outreach in one go, serving 47 small businesses per month and generating profit.
Frontend techniques
Enforce frontend code review and implementation quality using a frontend checklist
Use a structured frontend checklist and MCP tools to help AI agents and human developers conduct accessibility, performance, SEO, security, and code quality audits when writing React, Next.js, HTML, and CSS frontend code.
Equip AI with top-tier aesthetic judgment using 6 design and tool repositories
By integrating design system repositories, Figma MCP, and animation and 3D tool libraries, this solves the issues of ugly interfaces, missing animations, and a lack of professional design rules when AI writes frontend code. It enables AI to accurately translate Figma designs into code and generate animations and 3D scenes.
Ditching the AI look: use free, open-source design libraries to level up frontend UI quality
To fix the generic look common in AI-generated frontend interfaces, you can use Anthropic's official frontend design skills and open-source design libraries like UI UX Pro Max (featuring 79 UI styles) to eliminate the "AI look" through better taste, details, and motion effects.
Configuring rules to prevent AI from breaking your project
Set priorities, safety red lines, and modification scopes for AI coding assistants by configuring rule files like AGENTS.md or .cursor/rules/*.mdc, preventing unauthorized changes, hardcoded secrets, broken tests, and endless error loops.
Banish AI frontend aesthetic disasters with 7 design and animation toolkits
To address the common issue of visual ugliness or "AI slop" in AI-generated frontend interfaces, this article introduces 7 practical design libraries and tools covering color schemes, Figma-to-code conversion, animated components, and 3D generation to help AI tools shed their cheap look.
Three.js pixel art diorama scenes and QA workflows
Build pixel art diorama scenes in Three.js featuring locked orthographic cameras, color palette snapping, ordered dithering, and ink outlines using the Diorama playbook and code suite. Ensure visual fidelity and verifiable screenshots and animations through deterministic pure-function scene design and per-entity frame QA.
Ditch AI-generated frontend look with open-source design skills
Tailored for AI coding tools like Claude Code, Codex, and Cursor, use the Hallmark open-source design skill as an anti-AI trash design layer to prevent generating obvious AI-styled websites featuring default gradients, generic cards, and common SaaS layouts.
Reviewing AI-generated React Hooks code
AI-generated React hooks often look standard but hide subtle issues like deceptive useEffect dependencies, unstable callbacks, redundant derived state, and missing cleanup logic. This guide provides a five-point smell-test checklist for code reviews before merging AI-assisted branches.
Beautifying AI-Generated Frontends and Avoiding Common Pitfalls
AI-generated frontend interfaces often look cheap and lack polish. This card compiles practical advice from multiple designers, offering concrete improvement techniques and ready-to-use prompts covering everything from border-radius and color to typography and borders.
Cure the AI look in frontend code with these skills
Websites instantly look AI-generated when they lack dedicated styling skills. You can significantly improve frontend aesthetics by installing open-source skills and having your Agent rebuild the homepage.
Bypass AI frontend hallucinations using plugins and planning
Eliminate ugly AI-generated frontend designs, fabricated code, and non-compliant implementations by specifying explicit aesthetic styles, executing rigorous test-and-plan workflows, and cross-referencing the latest official documentation in real time.
Build high-quality 3D websites quickly with Claude
Generate working Three.js scenes in a single pass using refined prompts that strictly filter out useless output with Claude Code. The core technique is fast filtering rather than blindly generating numerous random versions, cutting a week of development down to a single afternoon.
Ditch cookie-cutter AI web design with the Hallmark skill package
To solve the cheap, generic look of AI-generated frontends—like oversized gradient headings and tacky neon accents—use the Hallmark (nutlope/hallmark) anti-homogenization design skill package. It features 48 physical material themes, a real open-source photography library, task-driven layouts, and one-click refactoring and design audits.
Design tips and standards for elevating AI frontend UI taste
Drawing on the insights of top designers, this guide helps you avoid cheap or tacky AI-generated interfaces by standardizing details like border radiuses, colors, typography, and borders. Effectively leveraging ready-made component libraries and prompts significantly enhances the UI quality in Vibe Coding.
Remove the generic AI look from frontend pages using Impeccable
Address common template-like and quality issues in AI-generated web pages using the open-source design optimization tool Impeccable. Supports multiple AI coding environments, from design reviews and rule checking to continuous iteration for higher delivery quality.
Feed real animation code to AI to skip endless ugly prompts
To solve the problem of AI generating ugly or rigid frontend pages, stop relying solely on long descriptive prompts. Instead, feed real single-file HTML code containing GSAP animations, complete design systems, or Rive-style interactions directly into AI coding tools so they can use it as a reference or refactor the page directly.
Using a Single-File HTML as Visual Reference to Reshape AI Frontend Appearance
When working with AI coding tools, instead of describing a design in vague language, provide a single-file HTML webpage that contains the complete visual language as reference. This allows the AI to extract and reuse the design system while preserving the application's existing functionality and business logic.
7 critical user journey tests for frontend deployment with AI assistance
AI coding tools can quickly generate working websites, but a few manual clicks don't prove real users can complete tasks. Before going live, you must test these 7 complete journeys from end to end: initial load, login/registration, core features, form submission, file upload, navigation routing, and permission sessions.
Escape generic AI-looking frontends using ux-skill
Install the ux-skill design intelligence engine (featuring 152 anti-pattern rules, 160 brand guidelines, and a 7-axis synthesizer) to inject deterministic constraints into AI coding tools like Cursor and Claude Code, ensuring generated UIs break away from cookie-cutter AI web aesthetics.
AI frontend and 3D development workflow for a 20-year backend veteran
A 20-year backend developer uses AI coding agents to build a 3D environment app for Vision Pro. By testing on real devices, explaining discomforts to the AI, and switching models to overcome roadblocks, they write or read zero lines of code.
Pitfalls and best practices for building 3D simulation frontends with AI
The author used AI (Opus) along with Three.js to build a 3D simulation frontend for the Singapore MRT. By reusing existing structured data, using artifacts for early visual decisions, and iterating on HUD design, you can significantly boost AI-driven frontend development efficiency.
Level up your frontend with free AI design resources
Collect and use free AI design resource sites to grab high-quality website prompts, backgrounds, animated blocks, and inspiration. Paste them directly into Claude or other AI tools to dramatically improve the quality of your frontend deliverables.
Collaborative 3D browser game development with AI
While recovering from illness, the author built the browser-based off-road racing game Slipface using Claude Code. By writing documentation, defining skills, providing reproduction links, and describing game feel, they achieved high-quality AI frontend and game development collaboration.
Fixing AI-generated frontends: eliminating fake hover states and layout clutter
The author hired a designer for $100 to conduct a UX audit on an AI-generated website. The original article outlines specific improvements for removing the AI-generated look, resolving conflicting CTAs, refining search, and handling static badge interactions.