Today's GitHub trending tells a clear story: the race to make coding agents smarter, more context-aware, and easier to configure has hit a new gear. A mix of individual contributors and established players are shipping tools that sit directly in the Claude Code and broader agentic IDE workflow — not just novel demos, but infrastructure that developers are adopting fast.
colbymchenry/codegraph: A Pre-Indexed Knowledge Graph for Your Codebase
codegraph picked up over 4,200 stars today — by far the biggest mover. It's a local-first tool that pre-indexes a codebase into a structured knowledge graph, making it queryable by Claude Code and other coding agents. Rather than forcing the agent to re-parse files on every request, Codegraph keeps a persistent, indexed representation of your code's symbols, relationships, and modules. For large codebases, this kind of persistent context layer can meaningfully reduce token overhead and improve retrieval accuracy.
github.com/colbymchenry/codegraph
multica-ai/andrej-karpathy-skills: One File of Prompting Wisdom
Inspired by Andrej Karpathy's public commentary on effective LLM usage, this single-file config distills prompting best practices into a Claude Code skill. It gained over 2,600 stars today, suggesting a lot of developers are looking for opinionated starting points rather than building their own from scratch. The appeal is simplicity — drop in one file and get measurably better coding agent behavior without writing custom tooling.
github.com/multica-ai/andrej-karpathy-skills
anthropics/claude-plugins-official: A Curated Plugin Directory
Anthropic shipped an official, maintained directory of high-quality Claude Code plugins, picking up nearly 700 stars. For a while, the Claude Code plugin ecosystem was fragmented — community plugins of varying quality scattered across GitHub. This repo centralizes vetted, well-maintained options in one place, lowering the friction of extending Claude Code for specific workflows. The Python repo is structured to be both browsable by humans and queryable by the agent itself.
github.com/anthropics/claude-plugins-official
obra/superpowers: A Methodology for Agentic Development
superpowers is a skills framework and software development methodology for agentic workflows, earning over 1,500 stars today. The project takes the position that effective use of coding agents isn't just about prompting — it requires a structured approach to breaking down work, handing off context, and verifying results. It comes as both a set of reusable skills and documentation of the philosophy behind them, making it part tooling and part playbook.
rohitg00/ai-engineering-from-scratch: A Hands-On Curriculum
ai-engineering-from-scratch is a structured learning curriculum for building real AI engineering projects, not just running tutorials. It gained over 1,300 stars and covers the full stack of what a practicing AI engineer does: data pipelines, fine-tuning, deployment, evaluation, and agents. The Python-based repo is organized as progressive hands-on projects rather than lectures, which is clearly resonating with developers who want to build intuition through doing rather than watching.