Recently updated open-source projects

Haptic is a local-first Markdown notes editor designed to keep notes and its database in the user's own environment. It includes a desktop application and a web version that can be self-hosted through Vercel or Docker. The project is still under active development: cross-device sync, note sharing, mobile apps, and Windows and Linux desktop support are listed as future work rather than current features.

IconChanger lets macOS users replace application and folder icons through either a graphical interface or a scriptable command-line tool. It integrates with macOSicons.com, previews changes in the Dock, and keeps favorites and history for reuse. A background service can restore custom icons on a schedule or after application updates overwrite them. The app is available through Homebrew or as a downloadable DMG.

claude-android-skill gives Claude Code an opinionated workflow for modern Kotlin applications, covering Jetpack Compose, MVVM and unidirectional data flow, layered architecture, offline-first storage, Hilt, and testing. Once copied into the skills directory, it activates for Android requests and can use bundled references, Gradle templates, and a feature-module generator. The repository is still marked as work in progress and is designed around this specific Android stack.

NeoDLP wraps yt-dlp in a cross-platform desktop interface for downloading individual media items or playlists from supported websites. Users can choose formats and streams, embed subtitles and metadata, apply SponsorBlock handling, configure proxies or rate limits, and send links from a Chromium or Firefox browser extension. Its packaged environment still relies on components such as yt-dlp, FFmpeg, Deno, and an optional Aria2 downloader that is not bundled with macOS builds. Users remain responsible for having permission to download the content.

media-stack assembles Radarr, Sonarr, Prowlarr, qBittorrent, Seerr, and Jellyfin into a self-hosted Docker Compose workflow for requesting, managing, and streaming media. An optional Recommendarr profile can generate film and TV suggestions from Sonarr, Radarr, Jellyfin, or Trakt data through a configured OpenAI API service, while download-related services can be routed through a VPN. Setup spans multiple containers, networks, permissions, and, for some VPN arrangements, manually assigned container addresses.

Modular Skills gives AI coding agents official, task-specific guidance for working with Mojo and the MAX platform. The collection can be installed all at once or skill by skill, with workflows for starting projects, handling Mojo–Python interoperability, importing model architectures, and debugging output divergence. It augments an existing coding agent rather than serving as a standalone development environment.

Lazyweb Skill Pack gives coding agents evidence-based design research before they build or critique an interface. Its local skills and hosted MCP can search real mobile and desktop screenshots, compare visual references, retrieve multi-screen journeys, guide screen design, map product flows, and propose reviewable changes. Installation and workflow discovery are broadly available, while data-bearing tools and limits depend on the account plan and experiment assignment. An optional autorouter edits only a marked instruction block after consent.

WebLLM Chat loads and runs language models inside a WebGPU-capable browser, keeping prompts, responses, and model processing on the user’s device instead of a cloud inference server. After the initial setup and model download, it can continue offline, with Markdown rendering, image-based conversations, and a selection of built-in models. Advanced users can also connect the interface to custom models served through MLC-LLM. Browser-native inference requires WebGPU support.

Zasper is a Jupyter-compatible IDE for data science and AI work that can run as a web or desktop application. It implements the Jupyter wire protocol, reuses installed kernels across languages, and combines notebook execution with an editor, terminal, launcher, and version-control workspace. The project emphasizes low resource use and many concurrent connections. macOS and Linux are fully supported, while Windows support is limited and the documentation recommends using WSL for the best experience.

Awesome Android Agent Skills packages Android architecture, Jetpack Compose, data-layer, accessibility, testing, and migration practices as installable Agent Skills. Copy the skills directory and shared Agent.md context into an Android project, then compatible assistants such as GitHub Copilot, Claude, Gemini, or Cursor can load the relevant instructions for each task instead of relying on repeated prompts.

LTX Desktop is an open-source desktop app for generating and editing videos with LTX models. It supports text-, image-, and audio-to-video generation, Retake editing, and project timelines. Windows or Linux systems with an NVIDIA CUDA GPU and at least 16 GB of VRAM can generate locally after downloading model weights; unsupported Windows hardware and macOS use the LTX API instead. The app is still in beta, local mode downloads large model files, and API-backed features send prompts and media to the service.

finance-quant-skills is a collection of Agent Skills focused on financial research and quantitative workflows, especially China's A-share market. Users can install all or selected skills for market data access, natural-language screening, backtesting frameworks, strategy documentation, and research-report generation. The repository is not a standalone trading terminal, and requirements vary by skill: some sources are free, while others need cookies, API tokens, or locally installed QMT or TongdaXin software.

OpenContext adds a personal knowledge store to existing coding-agent CLIs such as Codex, Claude Code, and OpenCode, so project background, decisions, and documents can persist across repositories and sessions. Users manage the library through the oc CLI, a desktop app, or a local web interface, while generated skills, slash commands, and an MCP server let agents search and update it directly. It reuses an agent CLI the user already has rather than bundling a separate model subscription.

Firebase Agent Skills is a collection of packaged instructions and scripts that help AI coding agents understand and work with Firebase. The bundle follows the open Agent Skills format and can be installed through the Agent Skills CLI, as a Gemini CLI extension, or through Claude and Codex plugin marketplaces. Manual copying is also documented for other compatible tools. It is not a standalone Firebase management application; its value depends on an agent environment that can load and apply the skills.

Turn a repository into a version-controlled n8n workspace where workflows can be pulled, edited, validated, diffed, and pushed from VS Code, Cursor, or the terminal. The extension, CLI, and Agent Skill share node schemas, examples, and live environment context, while TypeScript workflow files make changes easier for people and agents to review. It can also test executions, inspect credentials, and resolve sync conflicts, and works best with the latest stable n8n release.

Refly turns business procedures and team SOPs into reusable, versioned Agent Skills through a visual workflow builder. Skills can be installed, published, and shared through a registry, then exposed to other systems as APIs, webhooks, MCP services, or portable SDKs. It is aimed at teams that want durable agent capabilities rather than one-off prompts or copied automation templates.

PeterCat creates a repository-specific support bot from GitHub documentation and issue content. The bot can be embedded in a project website through its application SDK or installed as a GitHub App for repository workflows. A self-hosted deployment is available, but it requires several external services and credentials, including AWS, Supabase, authentication, language-model, and search components.

webclaw provides a local-first way to scrape, crawl, map, and extract websites for agents and RAG systems. Its core CLI and MCP workflow can run without an account for many pages, while a hosted API is available for bot-protected sites, JavaScript rendering, asynchronous crawl jobs, search, and production monitoring.