GitHub 2026-09-20 Daily Trending Projects

A coding-agent skill for security engineers that uses six phases and independent verification instead of one-shot scanning.

CUA provides cross‑platform sandboxes, unified SDKs, and benchmarks for AI agents that control full desktops, enabling local and cloud training, evaluation and deployment of real desktop tasks.

It gives AI coding agents an engineering workflow from requirements to release instead of just the shortest coding path.

Coder gives teams Terraform-managed cloud workspaces and AI agents while keeping model keys out of workspaces.

Claude Code is a terminal-based agentic coding assistant that understands your codebase via natural language, automates routine development tasks, explains complex code, and assists Git workflows—suited for efficiency-driven developers and teams.

OpenStock is a community open-source stock-tracking app built with Next.js/TypeScript offering charts, watchlists and email automation for self-hosting and education; limited by third-party market data availability and AGPL licensing.

A GPU and distributed-training manager for large-model teams, deploying Llama70b through GitHub workflows.

A Python tool that turns PDF and DOCX into generative-AI-ready structure, combining layout understanding, local execution, and framework integrations.

A low-level QUIC/HTTP/3 library for Rust and mobile apps where you own the network I/O and event loop.

Hister: self‑hosted private full‑text search for pages and local files, with optional semantic search.

A technology newsletter for developers, updated every Friday and broader than a single technical documentation set.

Everyone Can Use English is an open-source platform centered on AI-driven speech assessment and self-training, offering web and desktop clients for individuals and educators focused on long-term speaking and pronunciation improvement.

Role-focused customizable plugins to integrate tools and codify workflows for Claude, improving team consistency.

Needle 2 is an ultra‑compact model focused on tool calling and constrained decoding for structured extraction; a 14MB single file that enables offline edge inference and low‑resource deployment, though license and maintenance risks should be assessed.

For Codex desktop and CLI users, it puts prompts, providers, sessions, and MCP in one interface.