Recently updated open-source projects

NSFC Agent Skills offers four independent skills for structuring a proposal, creating research figures and roadmaps, managing literature and citations, and checking current funding policies. Researchers can load only the relevant skill in OpenClaw or another Agent Skills-compatible tool for each stage of the application. The collection is advisory: users must verify all generated material, follow research-integrity requirements, and defer to the official NSFC guidance for the applicable year.

VibeSkills is a workflow runtime that takes one request, turns it into bounded stages, and binds the relevant locally installed skills for planning, implementation, testing, documentation, research, or review. It records verification evidence and preserves workspace context for later sessions. The runtime is invoked through the host application's Skills entrypoint rather than as a standalone CLI.

JetBrains/skills collects 129 agent skills from 15 upstream repositories into a flat catalog that is easier to browse and reuse. Each skill preserves its original source link, while included CI can scan changed skills or audit the full repository. The collection is intended for demonstration and education, so critical workflows should still be tested independently.

Marketing Skills offers more than 160 Markdown-based playbooks for technical SEO, keyword research, page creation, paid ads, channels, launches, and analytics. You can install only the skills needed for a task in Cursor, Claude Code, OpenClaw, or other compatible agents. Tailored results depend on supplying project context such as the product, audience, brand, and target keywords.

Greptile Skills provides workflows for checking unresolved review comments and status failures, running a Greptile CLI review on the current checkout, and repeatedly reviewing and fixing a change. The skills can detect GitHub, GitLab, or Perforce from the environment, but you must install and authenticate the matching gh, glab, p4, or Greptile CLI first.

WG Tunnel is a desktop client for WireGuard and AmneziaWG configurations. Users can import, edit, export, sort, and monitor tunnels, while an independent lockdown mode blocks traffic when protection should remain enforced and can restore the previous tunnel at startup. The tunnel itself runs as a system service rather than depending on the graphical application, and configuration data is encrypted with system keychain integration. Current releases support Windows and Linux; Windows requires version 10.0.19041.0 or newer, while macOS support is still planned.

Ship Studio combines an AI agent terminal, live responsive preview, visual editing, Git branches and pull requests, conflict handling, and Vercel deployment in one local desktop workspace. It works with existing Claude Code, Codex, or OpenCode installations and keeps the repository on your machine, making it useful for developers, designers, freelancers, and agencies that want fewer tool switches. Users bring their own agent, GitHub, and hosting accounts. Official builds include anonymous usage analytics and crash reporting, both of which can be disabled in the application settings.

Row-Bot is a local-first desktop AI assistant for working across files, repositories, workflows, and messaging channels. It combines chat, durable memory, configurable tools, agent profiles, visible goals, and scoped child-agent delegation for longer tasks. Users can run models through Ollama or self-hosted endpoints, or opt into cloud and subscription providers. App data stays on the device by default, while content is sent externally only when a selected provider or integration requires it.

Qt AI Skills is the official agent-skill collection for Qt development and quality assurance. It covers Qt C++ and QML reviews, QML best practices, UI design, documentation generation, profiling, test creation and execution, Figma token and component conversion, and Qt 6 CMake projects. Selected review workflows combine deterministic linting with parallel AI analysis, and an official Qt documentation MCP server provides API lookup. The skills support Claude Code, Codex, Gemini CLI, and GitHub Copilot, but results still require careful review.

EvoSkill is an open-source framework for discovering and evolving reusable coding-agent skills. It runs agents on benchmark tasks, analyzes failed trajectories, proposes skill and system-prompt variants, and evaluates them on held-out data. Projects can use Claude Code, Codex, OpenCode, OpenHands, or Goose with CSV question-and-answer sets or containerized Harbor tasks. It suits teams with a meaningful evaluation set and requires the selected agent runtime, model-provider credentials, and supporting environment.

BrandDocs is an agent-skill bundle and Python CLI for generating new Word, PowerPoint, and Excel files from an organization's existing templates. It extracts a reusable brand profile covering themes, styles, structure, layouts, logos, tables, and spreadsheet rules, then builds new documents from the original template shell so formulas and native formatting remain intact. Each format stays separate rather than being converted into another. The workflow is local-first, but the full repository and Python dependencies are required; LibreOffice and Poppler are recommended for visual quality checks.

angular-electron is a starter project pairing Angular 22 with Electron 42 for desktop development, including a development server, renderer hot reload, Vitest and Playwright tests, and executable packaging for Linux, Windows, and macOS. Hot reload applies only to the renderer, so the Electron main process must restart, and the project explicitly recommends npm rather than Yarn.

MakoCode is a Windows desktop AI agent that wraps the Claude Code CLI in a visual-novel interface with character sprites, scenes, background music, voice assets, chat, and saves. Its first-run wizard detects Node.js, Git, and Claude Code, can install missing components in the background, and guides the user through testing a DeepSeek or other model-provider API configuration. The themed interface lowers setup friction, but the app still depends on Claude Code and a user-supplied model key.

software-design-philosophy is a reusable agent skill based on John Ousterhout's A Philosophy of Software Design. It gives coding assistants a design framework for code reviews, architecture discussions, API design, module decomposition, refactoring, naming, comments, and error-handling strategy. Users can install it in Claude Code, Codex, or compatible skill managers and invoke it alongside an existing project. The repository supplies guidance for making software-design decisions; it is not a standalone static analyzer or development application.

Platypus is a macOS developer tool that wraps shell, Python, Ruby, JavaScript, PHP, and other command-line scripts in native application bundles. The generated apps can show progress or output, accept dropped files, send notifications, run as status items, or work in the background. A graphical interface and command-line tool support both interactive packaging and automated builds.

Teams for Linux wraps the Teams web app in an unofficial desktop client with notifications, tray integration, screen sharing, custom backgrounds, and multiple account profiles. It is designed primarily for Linux and can be installed through common distribution packages or release builds. Some behavior remains limited by the Teams web experience, and the app disables Electron's built-in context isolation and sandboxing, so security-conscious users should consider running it through Flatpak, Snap, Firejail, or another system-level sandbox.

Add-to-Homescreen is a drop-in JavaScript library that shows platform-specific instructions for installing a website or PWA on iOS, Android, Windows, or macOS home screens. It supports more than twenty languages and can redirect users from several in-app browsers to the system browser. The site must already meet basic PWA requirements, including an app icon and web manifest.

Auto-Commenter is a Claude Skill workflow combining language-model generation with Playwright MCP for Reddit engagement. Users provide eight to ten of their own comments to create a style guide, configure target communities and optional product context, then let the workflow analyze posts, draft comments, fill daily quotas, record activity, and flag potential leads. Reddit is the implemented platform; other networks are extension patterns rather than finished integrations. Use still depends on following community rules, rate limits, and platform policies.