GitHub Trending Weekly Digest — April 28 to May 2, 2026

This week's GitHub Trending landscape reveals an explosive wave of AI agent infrastructure—from terminal reimagination to multi-agent orchestration frameworks. The common thread? Developers are moving beyond "AI that writes code" toward "AI that thinks like engineering teams."

🔥 Persistent Trendsetters (4+ Days on Trending)

mattpocock/skills

🔗 github.com/mattpocock/skills
Trending: 4 days (April 28, 29, 30, May 1)

What it does:
A battle-tested skill library for Claude Code and Codex, extracted from Matt Pocock's personal .claude directory. Instead of opinionated frameworks (GSD, BMAD), this provides composable, lightweight skills that embed software engineering best practices directly into AI workflows.

Why it matters:
AI agents accelerate code generation—but also accelerate entropy. Matt's solution: /grill-me for requirement alignment, /tdd for test-driven discipline, /improve-codebase-architecture for continuous design, and /diagnose for systematic debugging. Each skill is a distilled pattern from classics like The Pragmatic Programmer and Domain-Driven Design.

Tech:
Shell scripts, Claude Code/Codex skill system, designed for any LLM. Install with npx skills@latest add mattpocock/skills.

Key innovation:
The CONTEXT.md concept—a shared ubiquitous language that reduces token waste by ~75%. Instead of explaining "materialization cascade" every turn, the agent learns your project's vocabulary upfront.

---

🚀 Multi-Day Highlights (2-3 Days on Trending)

warpdotdev/warp

🔗 github.com/warpdotdev/warp
Trending: 3 days (April 29, 30, May 1)

What it does:
Warp reinvents the terminal as an agentic development environment. Built in Rust with OpenAI as a founding sponsor, it natively hosts coding agents (internal "Oz" agents or external CLI agents like Claude Code, Codex, Gemini CLI).

Why it matters:
Terminals haven't evolved for AI workflows. Warp's build.warp.dev lets you watch thousands of agents triaging issues, writing specs, implementing code, and reviewing PRs—all inside a web-compiled terminal UI.

Tech:
Rust, MIT-licensed UI framework (warpui_core), AGPL v3 for the rest. Leverages Tokio, NuShell, Alacritty, Hyper.

TauricResearch/TradingAgents

🔗 github.com/TauricResearch/TradingAgents
Trending: 3 days (April 30, May 1, 2)

What it does:
A multi-agent LLM framework simulating a trading firm's workflow: fundamental analysts, sentiment experts, technical analysts, research teams (bull vs. bear debate), traders, risk managers, and portfolio managers.

Why it matters:
Single AI models struggle with complex, multi-dimensional market analysis. TradingAgents orchestrates specialized agents via LangGraph, with persistent decision logs that inject historical performance (raw returns + alpha vs. SPY) into future runs—creating a reflexive learning loop.

Tech:
Python, LangGraph, multi-provider LLM support (OpenAI GPT-5.x, Gemini 3.x, Claude 4.x, Grok 4.x, DeepSeek, Qwen, GLM, OpenRouter, Ollama). Includes checkpoint recovery and backtesting.

ComposioHQ/awesome-codex-skills

🔗 github.com/ComposioHQ/awesome-codex-skills
Trending: 2 days (April 28, 29)

What it does:
A curated collection of 50+ Codex skills covering automation workflows: dev tools, collaboration, communication, data analysis, and meta-tools. Each skill is a modular SKILL.md with YAML metadata and progressive disclosure.

Why it matters:
Codex isn't just a text generator—it can send emails, create Linear issues, post to Slack, and operate 1,000+ apps via Composio CLI. This project packages proven workflows (e.g., brooks-lint for AI code review based on 6 engineering books, meeting-notes-and-actions for automated meeting transcripts).

Tech:
Python, Codex Skills framework, Composio CLI (1,000+ app integrations), MCP support. Install with python skill-installer/scripts/install-skill-from-github.py.

1jehuang/jcode

🔗 github.com/1jehuang/jcode
Trending: 2 days (April 29, May 1)

What it does:
A next-gen coding agent runtime built in Rust, designed for multi-session workflows, extreme customization, and performance. Features semantic memory, agent swarms (multi-agent collaboration in one repo), and Self Dev Mode (agents edit their own source code).

Why it matters:

• Performance brutality: 27.8 MB per session vs. Claude Code's 386.6 MB (13.9× lighter), 245× faster startup (14 ms vs. 3,437 ms)
• Native swarms: Multiple agents auto-resolve conflicts, broadcast messages, and coordinate tasks in the same codebase
• Human-like memory: Semantic vector embeddings + memory graphs auto-recall relevant context—no manual memory tool calls
• Self Dev Mode: Agents modify jcode's source, test changes, hot-reload, and continue working

Tech:
Rust, custom TUI (Handterm with 1,000+ fps rendering), custom Mermaid renderer (1,800× faster than browser-based), supports Claude/OpenAI/Gemini/Copilot/Ollama/vLLM.

soxoj/maigret

🔗 github.com/soxoj/maigret
Trending: 2 days (May 1, 2)

What it does:
An OSINT (Open Source Intelligence) tool that hunts for user profiles across 3,000+ websites using only a username—no API keys required. Extracts all available info (including linked accounts), supports recursive search, and bypasses censorship/CAPTCHA.

Why it matters:
Commercial OSINT tools (SocialLinks, UserSearch.ai) build on Maigret's MIT-licensed core. It offers multiple report formats (HTML, PDF, XMind mind maps, interactive D3 graphs), a built-in web UI, Tor/I2P support, and a Telegram bot for zero-install usage.

Tech:
Python 3.10+, socid_extractor, supports proxies/Tor/I2P, auto-updates site database from GitHub every 24 hours.

---

💎 Notable Single-Day Appearances

abhigyanpatwari/GitNexus

🔗 github.com/abhigyanpatwari/GitNexus
Trending: April 28

What it does:
A zero-server code intelligence engine that indexes any codebase as a knowledge graph (dependencies, call chains, clusters, execution flows) and exposes it to AI agents via MCP (Model Context Protocol).

Why it matters:
Traditional AI coding assistants lack structural awareness—they don't know that 47 functions depend on UserService.validate()'s return type. GitNexus precomputes relationships with smart clustering, tracing, and scoring. One query returns complete context.

Tech:
Node.js native + Tree-sitter (14 languages) + LadybugDB (embedded graph database) + BM25 + semantic vectors + RRF hybrid search + MCP (16 tools). Includes impact analysis ("blast radius" before changes), Wiki generation, and Claude Code PreToolUse hooks for auto-enhanced search.

microsoft/VibeVoice

🔗 github.com/microsoft/VibeVoice
Trending: April 28

What it does:
Microsoft's frontier speech AI family: TTS (text-to-speech) and ASR (automatic speech recognition) models using 7.5Hz ultra-low-frame-rate continuous speech tokenizers + next-token diffusion framework.

Why it matters:

• VibeVoice-ASR-7B: Processes 60-minute audio in one pass (no chunking), structured transcription (Who/When/What), 50+ languages, custom hotwords
• VibeVoice-TTS-1.5B: Generates 90-minute long-form speech with 4 different speakers, expressive dynamics, spontaneous singing
• VibeVoice-Realtime-0.5B: Real-time TTS (~300ms latency), streaming text input, 11 English styles + experimental multilingual voices

Tech:
7.5Hz continuous speech tokens, Qwen2.5 1.5B LLM backbone, diffusion head for acoustic fidelity, ICLR 2026 Oral paper.

davila7/claude-code-templates

🔗 github.com/davila7/claude-code-templates
Trending: April 28

What it does:
A ready-to-use collection of 100+ Claude Code configs: agents, custom commands, settings, hooks, MCP integrations, and project templates. Includes a CLI browser and real-time analytics dashboard.

Why it matters:
Claude Code is powerful but complex to configure. This project provides categorized templates (security auditors, React performance optimizers, database architects) + web dashboard (Beta) + health checks + analytics.

Tech:
Node.js, Claude Code config system, React dashboard (Beta), Cloudflare Tunnel for secure remote access. Sources include K-Dense-AI scientific skills (139), Anthropic official skills (31), community contributions.

HunxByts/GhostTrack

🔗 github.com/HunxByts/GhostTrack
Trending: April 29

What it does:
A practical OSINT toolkit for tracking IP geolocation, phone number attribution, and username activity across social media platforms.

Why it matters:
Provides three modules: IP Tracker (pairs with Seeker tool), Phone Tracker (carrier/location lookup), and Username Tracker (social media footprint). Version 2.2 with active development.

Tech:
Python 3, supports Linux (apt) and Termux (pkg).

obra/superpowers

🔗 github.com/obra/superpowers
Trending: April 30

What it does:
A complete agent software development methodology + composable skill framework. Enforces TDD, YAGNI, DRY principles with automated workflows.

Why it matters:
Implements a structured pipeline: brainstorming (Socratic questioning) → git-worktrees (branch isolation) → writing-plans (2-5 minute tasks) → subagent-driven-development (two-stage review: spec compliance + code quality) → test-driven-development (RED-GREEN-REFACTOR) → requesting-code-review (inter-task inspection) → finishing-a-development-branch (merge/PR/keep/discard).

Tech:
Cross-platform skill framework supporting Claude Plugin Market, superpowers-marketplace, Gemini/OpenCode extensions, Cursor/Copilot.

lukilabs/craft-agents-oss

🔗 github.com/lukilabs/craft-agents-oss
Trending: April 30

What it does:
An Agent Native desktop application combining Claude Agent SDK and Pi SDK. Provides multi-session inbox, seamless API/MCP integration, headless server mode, and automation system.

Why it matters:

• Multi-session inbox: Todo → In Progress → Needs Review → Done workflow
• Natural language API setup: "add Linear as a source" → agent auto-configures
• Multi-provider: Claude (API Key/OAuth), Google AI Studio, ChatGPT Plus/Pro, GitHub Copilot, OpenRouter, Ollama
• Headless server mode: Remote VPS deployment with WebSocket/TLS, thin desktop client
• CLI tools: craft-cli ping/health/sessions/send/run with 21-step validation (--validate-server)
• Automation: Event-driven (LabelAdd, SchedulerTick, PreToolUse), cron scheduling, prompt actions

Tech:
TypeScript, Electron, React, Bun, shadcn/ui, Tailwind CSS v4, AES-256-GCM encrypted storage, Docker support.

ruvnet/ruflo

🔗 github.com/ruvnet/ruflo
Trending: May 2

What it does:
A leading agent orchestration platform for Claude, supporting intelligent multi-agent swarms, autonomous workflows, and conversational AI systems.

Why it matters:
Adds a "nervous system" to Claude Code: agents self-organize into swarms, learn from tasks, maintain cross-session memory, and communicate federatively (secure cross-machine collaboration). Includes 100+ specialized agents, distributed swarm intelligence, RAG integration, and 32 native Claude Code plugins.

Tech:
TypeScript/JavaScript, Rust (WASM kernel), LangChain, Claude/GPT/Gemini/Cohere/Ollama, WebAssembly, Docker. Features zero-trust federation, SONA neural patterns, ReasoningBank, AgentDB (HNSW-indexed vector memory, 150×-12,500× faster search).

Extras:

• Web UI Beta at flo.ruv.io: Multi-model chat + parallel MCP tool calling
• GOAP planner at goal.ruv.io: Natural language goals → executable agent plans

browserbase/skills

🔗 github.com/browserbase/skills
Trending: May 2

What it does:
Claude Agent SDK with web browsing tools, enabling Claude Code to collaborate with Browserbase for browser automation.

Why it matters:
AI agents need to access dynamic web pages, handle CAPTCHAs, and bypass bot detection. This SDK provides skills (browser, browserbase-cli, functions, site-debugger, browser-trace, cookie-sync, fetch, search, ui-test) for:

• Automated browser interactions (remote Browserbase sessions with anti-bot stealth, CAPTCHA solving, residential proxies)
• Serverless browser automation deployment to Browserbase cloud
• Diagnosing/fixing failed automations
• AI-driven adversarial UI testing

Tech:
JavaScript, Browserbase Functions, Playwright, Chrome CDP, bb CLI. Includes full DevTools Protocol tracing (CDP firehose, screenshots, DOM dumps), cookie sync from local Chrome, and AI-analyzed git diffs for bug detection.

D4Vinci/Scrapling

🔗 github.com/D4Vinci/Scrapling
Trending: May 2

What it does:
An adaptive web scraping framework handling everything from single requests to full-scale crawls. Parsers learn from website changes and auto-relocate elements when pages update.

Why it matters:
Traditional scrapers break when sites deploy anti-bot systems (Cloudflare Turnstile) or change structure. Scrapling offers:

• Out-of-the-box Cloudflare Turnstile bypass
• Intelligent element tracking: similarity algorithms re-locate elements after site changes
• Unified interface: HTTP requests, stealth browsers, dynamic automation
• Scrapy-style crawler API: concurrent crawling, multi-session support, pause/resume

Tech:
Python 3.10+, Playwright, Chromium, Lxml, HTTP/3. Performance: ~784× faster than BeautifulSoup, ~12× faster than PyQuery. Includes built-in MCP server for AI-assisted scraping (Claude/Cursor), interactive IPython shell, streaming mode (async for item in spider.stream()).

---

🎯 Weekly Themes

AI Agent Infrastructure Maturity

• Terminal reimagination (Warp)
• Multi-agent orchestration (Ruflo, TradingAgents)
• Performance engineering (jcode: 13.9× lighter memory, 245× faster startup)
• Engineering discipline automation (mattpocock/skills, obra/superpowers)

Code Intelligence & Context Awareness

• GitNexus: Knowledge graphs for codebases
• jcode: Semantic memory systems
• mattpocock/skills: CONTEXT.md ubiquitous language

Web Automation & Scraping Renaissance

• Browserbase/skills: AI-driven browser automation
• Scrapling: Adaptive, anti-bot scraping
• maigret: OSINT at scale (3,000+ sites)

Voice AI Breakthroughs

• Microsoft VibeVoice: 60-min single-pass ASR, 90-min TTS with spontaneous singing

Cross-Platform & Multi-Provider Ecosystems

• Craft Agents: Claude/Google/ChatGPT/Copilot unified desktop app
• awesome-codex-skills: 1,000+ app integrations via Composio
• jcode: Claude/OpenAI/Gemini/Ollama/vLLM support

---

💡 Key Takeaways

1. "AI that thinks like teams" is the new frontier—multi-agent swarms (Ruflo, TradingAgents, jcode) are replacing monolithic models

2. Engineering discipline matters more than ever—Matt Pocock's skills and obra's superpowers embed TDD, DDD, and systematic debugging into AI workflows

3. Performance is a feature—jcode proves extreme optimization is possible (27.8 MB vs. 386.6 MB), forcing incumbents to rethink architecture

4. Context is king—from GitNexus's code knowledge graphs to mattpocock's CONTEXT.md, tools that preserve and leverage structural context win

5. The terminal is dead, long live the agentic terminal—Warp's reimagination signals a broader shift from "command executor" to "AI collaboration canvas"

6. Cross-provider neutrality is table stakes—every major tool now supports Claude, OpenAI, Gemini, and local/self-hosted models

7. OSINT + web automation are AI superpowers—maigret, Scrapling, and browserbase/skills show how AI unlocks previously manual intelligence gathering

---

Compiled by Tommy Zhang | May 3, 2026