GitHub Trending Weekly Digest — Feb 23-28, 2026

Welcome to this week's GitHub Trending digest! The open-source world has been buzzing with innovation, and this week's highlights show a clear trend: AI agent infrastructure is maturing at breakneck speed. From context engineering to multi-agent orchestration, developers are building the foundation for truly autonomous coding assistants.

🔥 Trending Champions: Multi-Day Winners

These projects dominated the trending charts across multiple days, proving their staying power and developer interest.

Agent Skills for Context Engineering (5 days)

🔗 github.com/muratcankoylan/Agent-Skills-for-Context-Engineering

What it does: A comprehensive skill collection focused on context engineering principles for building production-grade AI agent systems. Covers multi-agent architecture, memory systems, tool design, and evaluation frameworks.

Why it matters: Context window management is the core challenge of AI agents. As context grows, models suffer from "lost-in-the-middle" phenomena and U-shaped attention curves. This repo systematically teaches how to curate high-signal context—system prompts, tool definitions, retrieved documents, message history—to maximize agent effectiveness. It's been cited by Peking University's General Artificial Intelligence Lab as foundational work for static skill architecture.

Tech: Python, Claude Code Plugin, BDI cognitive architecture, LLM-as-Judge, RDF/OWL ontologies, progressive disclosure, modular isolation

Hugging Face Skills (4 days)

🔗 github.com/huggingface/skills

What it does: Standardized skill definitions for AI/ML tasks (dataset creation, model training, evaluation) compatible with all major coding agents—Claude Code, OpenAI Codex, Gemini CLI, and Cursor.

Why it matters: Different AI coding tools use different instruction formats, forcing developers to rewrite configurations for each platform. Hugging Face Skills adopts a standardized Agent Skill format (SKILL.md + YAML frontmatter), making the same skill definition work across platforms. Each skill is a self-contained folder with instructions, scripts, and resources, covering Gradio UI building, Hub operations, dataset management, model training (TRL, SFT, DPO, GRPO), evaluation (vLLM, lighteval), and paper publishing.

Tech: Python, Hugging Face Hub CLI, YAML, Gradio, TRL, vLLM, lighteval, Trackio, Modal cloud compute, GGUF conversion

Superpowers (3 days)

🔗 github.com/obra/superpowers

What it does: A complete software development workflow framework for AI coding agents, built on composable "skills" that give your code assistant superpowers.

Why it matters: Traditional code assistants jump straight into writing code without systematic planning and validation, leading to implementations that drift from requirements. Superpowers establishes a full development workflow: start with dialogue to clarify needs and generate specifications, get user confirmation, then create a detailed implementation plan (emphasizing TDD, YAGNI, DRY). It uses subagent-driven development (SDD) to autonomously execute tasks, including two-stage reviews (spec compliance + code quality). Uses Git worktrees to isolate dev branches, and automatically validates and offers merge/PR options when tasks complete. Includes 7 core skills (brainstorming, planning, TDD, systematic debugging, code review) and 4 principles (test-first, systematic over ad-hoc, complexity reduction, evidence over assertion).

Tech: Shell, Git worktrees, Subagent orchestration, RED-GREEN-REFACTOR TDD, SKILL.md spec, Claude Code/Cursor/Codex/OpenCode plugin marketplace

📈 Multi-Day Risers: 2-Day Trending

OpenBB

🔗 github.com/OpenBB-finance/OpenBB

What it does: Financial data platform for analysts, quant researchers, and AI agents. Provides "connect once, use everywhere" open-source tooling that integrates proprietary, licensed, and public data sources.

Why it matters: Financial data is scattered across multiple sources (Bloomberg, Reuters, public APIs). Analysts and developers have to integrate repeatedly. OpenBB acts as an infrastructure layer, unifying data access and exposing it to multiple applications: Python environments (quant research), OpenBB Workspace and Excel (analysts), MCP servers (AI agents), and REST APIs (other apps).

Tech: Python (3.9-3.12), FastAPI, Uvicorn, MCP (Model Context Protocol), Excel integration, REST API

system-prompts-and-models-of-ai-tools

🔗 github.com/x1xhlol/system-prompts-and-models-of-ai-tools

What it does: A complete collection of system prompts, internal tools, and AI models from 30+ AI tools (Augment Code, Claude Code, Cursor, Devin AI, Perplexity, etc.)—over 30,000 lines of detailed insights.

Why it matters: AI startups and developers often need to understand how other AI products architect their systems and design prompts, but this information is typically not public. This project helps developers learn best practices while also warning AI startups about data security (exposed prompts and models can become hacker targets).

Tech: Open-source data collection, security auditing, prompt engineering

deer-flow (ByteDance)

🔗 github.com/bytedance/deer-flow

What it does: Open-source super-agent framework with research, coding, and creation capabilities, handling tasks from minutes to hours.

Why it matters: DeerFlow evolved from a deep research framework into a super-agent runtime environment. Traditional agents lack infrastructure support for complex multi-step tasks. DeerFlow 2.0 (rewritten from scratch on LangGraph and LangChain) provides out-of-the-box file systems, memory, skill libraries, sandboxed execution, and subagent orchestration. Supports progressive skill loading (only load when needed to keep context tight), parallel subagent execution, cross-session long-term memory. Each task runs in an isolated Docker container with full filesystem and bash execution.

Tech: TypeScript, LangGraph, LangChain, Docker, Kubernetes, MCP Servers, Skill-based architecture, Sandboxed execution

💎 One-Day Stars: Fresh Innovation

Scrapling

🔗 github.com/D4Vinci/Scrapling

What it does: An adaptive web scraping framework that handles everything from single requests to full-scale crawlers.

Why it matters: Traditional scrapers face two pain points: selectors break when sites update, and anti-bot systems (like Cloudflare Turnstile) are hard to bypass. Scrapling's parser learns from site changes and automatically re-locates elements when pages update; its fetcher bypasses anti-bot systems out of the box. The crawler framework supports pause/resume, automatic proxy rotation—all with just a few lines of Python. Includes MCP server for AI-assisted scraping, plus interactive shell and CLI tools.

Tech: Python, Tree-sitter (AST parsing), HTTP/3, Playwright/Chrome automation, KuzuDB (embedded graph DB), transformers.js (browser ML), Docker, asyncio

GitNexus

🔗 github.com/abhigyanpatwari/GitNexus

What it does: Zero-server code intelligence engine—indexes any codebase as a knowledge graph in your browser, including dependencies, call chains, clusters, and execution flows, exposing them to AI agents via smart tools.

Why it matters: AI coding tools like Cursor and Claude Code are powerful but lack deep structural understanding of codebases, easily missing dependencies, breaking call chains, or blindly committing changes. GitNexus indexes codebases as knowledge graphs (tracking every dependency, call chain, cluster, execution flow) and exposes them to AI agents via MCP protocol. Provides 7 MCP tools (hybrid search, 360° symbol view, impact analysis, change detection, multi-file rename, Cypher queries) and 4 agent skills (explore, debug, impact analysis, refactor). CLI indexes local repos, Web UI runs entirely in-browser (client-side, no server). Pre-computed relationship intelligence means agents get complete context in one call, avoiding multi-round queries.

Tech: TypeScript, Tree-sitter, KuzuDB, MCP, Sigma.js (WebGL graph rendering), transformers.js, Graphology, React 18, Vite, Tailwind v4, BM25 + semantic search + RRF, WebAssembly, Web Workers + Comlink

SpacetimeDB

🔗 github.com/clockworklabs/SpacetimeDB

What it does: A relational database that merges database and server into one, letting clients connect directly to execute application logic.

Why it matters: Traditional architecture requires separately deploying web/game servers, databases, microservices, container orchestration—complex to operate. SpacetimeDB lets you upload application logic directly to the database as "modules" to run, with clients connecting directly to execute business logic. Your entire app can be written in Rust and deployed as a single binary—no microservices, Kubernetes, or Docker needed. Real-time applications (like the MMORPG BitCraft Online) run all backend logic in a single SpacetimeDB module.

Tech: Rust, C#, TypeScript, WAL (Write-Ahead-Log), In-Memory Database, Docker, BSL 1.1 License (converting to AGPL v3.0 with linking exception)

moonshine (moonshine-ai)

🔗 github.com/moonshine-ai/moonshine

What it does: Fast, high-accuracy automatic speech recognition (ASR) system optimized for edge devices and real-time voice applications.

Why it matters: Traditional ASR models like Whisper require fixed 30-second input windows, can't cache, have high latency (seconds), and poor multilingual support. Moonshine is designed for real-time voice interfaces: supports flexible-length input, streaming cache (810× latency reduction), language-specific models (higher accuracy), cross-platform unified API (iOS/Android/Windows/Linux/Pi). The Medium Streaming model has only 245M parameters but outperforms Whisper Large v3 (1.5B params), with just 107ms latency on MacBook Pro (vs Whisper's 11,286ms). Provides complete transcription, speaker identification, and intent recognition APIs.

Tech: C, C++, Python, Swift, Java, Rust, OnnxRuntime, CMake, speech recognition, streaming processing

wifi-densepose

🔗 github.com/ruvnet/wifi-densepose

What it does: WiFi signal-based human pose estimation system—no cameras needed, uses regular routers for real-time full-body tracking through walls.

Why it matters: Traditional pose estimation relies on cameras, raising privacy concerns and unable to penetrate obstacles. This project uses WiFi Channel State Information (CSI) and machine learning for camera-free pose detection: supports real-time processing (<50ms latency, 30 FPS), multi-person tracking (up to 10), through-wall detection. Use cases include medical monitoring (fall detection), smart homes, disaster rescue (detecting survivors in rubble, vital sign detection). Provides REST API, WebSocket streaming, 100% test coverage. Rust implementation achieves 810-5400× performance improvement.

Tech: Python, Rust, FastAPI, PyTorch, Docker, Kubernetes, WiFi CSI, neural networks, WebSocket

AIRI (moeru-ai)

🔗 github.com/moeru-ai/airi

What it does: Self-hosted AI virtual companion/digital life container, inspired by Neuro-sama, supports real-time voice chat, plays Minecraft/Factorio, runs on Web/macOS/Windows.

Why it matters: Existing AI chat platforms (like Character.ai) only handle text interaction, can't play games or engage in multimodal interaction like Neuro-sama. AIRI gives you a true "digital life"—not just chatting, but gaming together, watching videos, 24/7 companionship.

Tech: TypeScript, Vue.js, WebGPU, WebAudio, WebAssembly, Tauri, Live2D/VRM, DuckDB WASM, ElevenLabs TTS, LLM integration (OpenAI/Claude/Gemini/local models)

Claude Code (Anthropic official)

🔗 github.com/anthropics/claude-code

What it does: Official smart coding tool from Anthropic, runs in terminal, understands codebases, helps execute daily tasks, explains complex code, handles Git workflows via natural language commands.

Why it matters: Developers constantly switch between terminals and codebases, manually executing repetitive tasks (like Git operations, code reviews) inefficiently. Claude Code lets developers use natural language to directly command AI to complete this work, dramatically boosting coding speed.

Tech: Node.js 18+, Claude AI, plugin system, Git integration, cross-platform support (macOS/Linux/Windows)

XZ Utils

🔗 github.com/tukaani-project/xz

What it does: General-purpose data compression library and command-line tool, supports .xz and .lzma formats, LZMA2 algorithm provides 30% better compression than gzip, widely used for software distribution.

Why it matters: Software package distribution needs to balance compression ratio, speed, and decompression performance. XZ Utils provides higher compression than gzip (saving storage and bandwidth) while decompressing faster than bzip2, suitable for scenarios requiring multiple decompressions (like Linux distro packages).

Tech: C, LZMA2 compression algorithm, GNU Autotools, multi-threaded compression, zlib-compatible API

Awesome LLM Apps

🔗 github.com/Shubhamsaboo/awesome-llm-apps

What it does: Curated collection of LLM applications, including RAG, AI Agents, multi-agent teams, MCP, voice agents, supporting OpenAI/Anthropic/Gemini/xAI and open-source models.

Why it matters: Developers want to build practical applications with LLMs but don't know where to start, lacking complete reference examples. This repo provides 100+ real working projects (from travel planning, financial analysis to game AI), helping developers quickly learn and get started.

Tech: Python, OpenAI/Anthropic/Gemini/Llama APIs, RAG (vector databases), AI Agents (Phidata/CrewAI), MCP, Voice AI (ElevenLabs), multi-agent frameworks

Ruflo

🔗 github.com/ruvnet/ruflo

What it does: Leading Claude agent orchestration platform, supports deploying multi-agent clusters, coordinating autonomous workflows, building conversational AI systems with enterprise architecture, distributed swarm intelligence, RAG integration, and native Claude Code/Codex integration.

Why it matters: Single AI agents have limited capabilities; complex tasks require multiple specialized agents collaborating. Ruflo provides enterprise-grade agent orchestration, letting you conduct an AI team like an orchestra—automatically assigning tasks, coordinating workflows, integrating results.

Tech: TypeScript, Claude AI, multi-agent systems, Swarm Intelligence, RAG integration, MCP Server, Claude Code skill system, Kubernetes/Docker runtime

prompts.chat

🔗 github.com/f/prompts.chat

What it does: World's largest open-source AI prompt library (formerly Awesome ChatGPT Prompts), 143k+ GitHub stars, supports ChatGPT, Claude, Gemini, Llama, Mistral, and all major AI models.

Why it matters: Users often don't know how to write effective AI prompts to elicit models' best performance. prompts.chat provides community-curated prompt collections, interactive prompt books (25+ chapters covering basics to advanced techniques), kids' coding games (ages 8-14), self-hosted solutions (GitHub/Google/Azure AD auth), CLI tools, Claude Code plugin, and MCP server integration. Featured by Forbes, cited by Harvard and Columbia, most-liked dataset on Hugging Face.

Tech: Next.js, MCP server, Claude Code plugin, Docker, npx CLI, CSV/Markdown data formats

LadybirdBrowser

🔗 github.com/LadybirdBrowser/ladybird

What it does: A truly independent web browser built with a new engine based on web standards, using multi-process sandboxed architecture for security.

Why it matters: Breaks the current browser market monopoly by a few engines (Chromium, WebKit, Gecko), providing a completely independent browser engine implementation. Multi-process architecture (UI process, WebContent rendering process, ImageDecoder process, RequestServer process) and sandbox isolation enhance browser security and robustness.

Tech: C++, LibWeb (Web rendering engine), LibJS (JavaScript engine), LibWasm (WebAssembly), LibCrypto/LibTLS, LibHTTP, LibGfx, Multi-process architecture, Sandboxing

🎯 Weekly Themes

AI Agent Infrastructure Dominance

Context Engineering & Memory Systems
Multi-Agent Orchestration
Standardized Agent Skills
Production-Grade Agent Frameworks

Developer Tools Evolution

Code Intelligence & Knowledge Graphs
Automated Workflows & TDD
Real-Time ASR for Voice Interfaces

Breaking Monopolies

Independent Browser Engines
Unified Financial Data Access
Open-Source Prompt Libraries

Emerging Tech Frontiers

WiFi-Based Pose Estimation
AI Virtual Companions
Database-as-Runtime Architecture

💡 Key Takeaways

This week painted a clear picture: AI agent development is graduating from experimentation to production. The simultaneous trending of context engineering (Agent-Skills-for-Context-Engineering), standardized skills (Hugging Face Skills), and complete workflows (Superpowers) shows that developers are no longer satisfied with basic autocomplete—they want AI assistants that can autonomously handle multi-hour tasks while maintaining quality and staying on-track.

The emergence of knowledge graph-based code intelligence (GitNexus) and super-agent orchestration platforms (DeerFlow, Ruflo) suggests we're building the infrastructure layer for truly autonomous coding. These aren't just tools; they're the foundation for the next generation of software development.

Meanwhile, innovations in privacy-preserving tech (WiFi pose estimation), edge AI (Moonshine ASR), and independent infrastructure (LadybirdBrowser, SpacetimeDB) remind us that open source thrives when it challenges the status quo.

Weekly digest curated from GitHub Trending (Feb 23-28, 2026). Total unique projects: 18. Data compiled and analyzed by Tommy Zhang.