13 November 2026

Modern web practices for coding agents

Modern Web Guidance is a set of skills with web platform best practices and browser compatibility data for coding agents. This repo is the source for authoring, calibrating, and evaluating its contents.

• Agents often default to older patterns because LLM training data is full of legacy code, and these skills steer them to modern web solutions.
• The search, retrieve, and legacy-scan commands work without installing, to find guides or scan a codebase for legacy patterns.
• Stewarded by the Chrome, Edge, Firefox, and MDN teams with the web development community.

OpenAI CLI and SDK for finding and fixing vulnerabilities

Codex Security scans repositories, validates findings, generates patches, and verifies fixes. It runs in Codex, through the cs CLI, or from the TypeScript SDK and GitHub Action.

• Scans can cover a whole repo, selected paths, or Git changes, in Standard and Deep modes.
• Findings can be tracked in Linear, Jira, or GitHub, and exported to SARIF, JSON, and CSV.
• Requires Node.js 22.13.0 or later within 22.x, or Node.js 24.

Node.js version manager for POSIX shells

nvm is a bash script to install and switch between multiple active Node.js versions. It is installed per user and invoked per shell, in sh, dash, ksh, zsh, and bash.

• One command such as nvm install 24 or nvm use 22 installs or switches the Node.js and npm versions.
• Works on Unix, macOS, and Windows WSL.
• A .nvmrc file and shell integration let nvm use run automatically in a directory.

Enable Hermès watch faces on Apple Watch

HermesEnabler enables Hermès watch faces on any Apple Watch. It is supported only on watchOS 27.0 through 27.3 beta 1, and the iPhone must be paired with the target Watch.

• Install the prebuilt IPA with Sideloadly, or build the app directly with Xcode.
• The README notes that Sideloadly 0.72 may remove the Watch app while signing, and suggests Xcode if that happens.
• Developer Mode must be on for both the iPhone and the Apple Watch before opening the app.

Open source personal finance app, a fork of Maybe

Sure is a community-maintained personal finance app, a fork of the abandoned Maybe Finance project. It can be self-hosted with Docker.

• Reachable from a browser, the macOS desktop app, the mobile app, API clients, and LLM agents.
• Maybe Finance stopped development in mid-2023 after its hosted version did not hold up as a business.
• Licensed AGPLv3, and the team asks contributors to join this repo rather than fork it.

Open source coding agent for terminal, editor, and chat

Qwen Code is an open source coding agent that runs in your terminal, with IDE plugins, a desktop app, Web UI, SDKs, and chat integrations.

• Supports OpenAI, Anthropic, Gemini, and Qwen APIs, plus local models through Ollama or vLLM, switchable at runtime.
• Agent features include Auto-Memory, Auto-Skills, SubAgents, Agent Teams, and MCP.
• Chat integrations are available for Telegram, DingTalk, WeChat, and Feishu.

Zig LLM inference engine for Metal and CUDA

TensorFold serves language models from a single Zig binary on Apple Silicon and NVIDIA GPUs. The engine reads checkpoints, tokenizes requests, and runs Metal or CUDA kernels directly.

• Serving needs no Python or MLX installation.
• tensorfold pull downloads a model and tensorfold serve starts a server reachable through the /v1/chat/completions endpoint.
• Contributors and agents are asked to use Agent Traffic Control (ATC) to avoid duplicate work on an issue.

Open source cross-platform JavaScript runtime

Node.js is an open source, cross-platform JavaScript runtime. The project uses an open governance model and is supported by the OpenJS Foundation.

• Each October a new release line branches off main and gets Alpha releases for 6 months, then is promoted to Current each April.
• Nightly releases and per-version API documentation are also available.
• Binaries are verified with SHASUMS256.txt.asc, which holds SHA checksums and the releaser PGP signature.

Engineering skills for AI coding agents

Agent Skills packages the workflows and quality gates senior engineers use as skills, so AI coding agents follow them in every development phase. It ships 9 slash commands for the lifecycle.

• The commands are /spec, /plan, /build, /test, /constraints, /review, /webperf, /code-simplify, and /ship.
• /build auto generates the plan and implements every task in one pass after you approve the plan, with each task still test-driven and committed individually.
• In Claude Code the commands install as plugin commands and use a namespaced form, such as /agent-skills:review.

Open source design system from Meta for React

Astryx is an open source design system that grew inside Meta over eight years, with 150+ accessible components, theming, dark mode, templates, and a CLI. People and AI assistants use the same tooling.

• Uses pre-built CSS and typed React components, with no build plugin or new styling library to adopt.
• A theme is a set of CSS custom property overrides, and swizzle ejects a component's source into your project.
• Requires React 19 or later, and is currently in Beta.

Self-hostable AI coworkers with approval for every action

OpenBot is a modular workspace for AI coworkers, each with its own computer holding a browser, files, and tools. Every action is decided before it happens and recorded after.

• Bring any AG-UI agent, choose your own models, and run it on your own infrastructure.
• A bot waits for your approval before it sends something, and you can take over when it needs you.
• A laptop app (Mac and Windows) with a Docker-based backend is available, plus OpenBot for Business for teams.

Router that connects coding agents to any model provider

free-claude-code routes supported coding harnesses to models from supported providers. It is an independent project not affiliated with Anthropic.

• Supports 12 harnesses, including Claude Code, Codex, Pi, OpenCode, Cline, and Aider, and 61 free, paid, subscription, and local providers.
• The README cites 1.3B+ free tokens per month, and integrations are removed if a provider stops allowing them.
• Usable from the terminal, desktop apps, and the VS Code and JetBrains IDE extensions.

LLM inference in C/C++

llama.cpp runs LLM inference in C/C++. It can download and run a model straight from Hugging Face, or launch an OpenAI-compatible API server.

• llama cli -hf ggml-org/Qwen3.5-0.8B-GGUF downloads a model from Hugging Face and runs it directly.
• llama serve launches an OpenAI-compatible API server.
• Installed with an install script, or PowerShell on Windows.