Top 10 Trending GitHub Repos This Week (September 9 – September 15, 2026)

TL;DR: This week’s trending list is dominated by AI-agent tooling — skills, harnesses, and CLI utilities built specifically to make Claude Code, Codex, Cursor, and similar coding agents faster, less wasteful, and less prone to over-engineering. Four of the ten repos are JavaScript-based, and several are direct responses to pain points anyone running AI coding agents daily will recognize: buried answers, blown context windows, and agents that turn simple fixes into frameworks.

Here are the 10 repositories that trended on GitHub between September 9 and September 15, 2026. As usual, this roundup skips straight past the marketing copy and tells you what each project actually does, why it’s catching attention right now, and who should actually go take a look.

1. ayghri/i-have-adhd

github.com/ayghri/i-have-adhd — Python, 46,887 stars

This is a “skill” (a packaged instruction set you can plug into an AI coding agent) that stops your agent from burying the actual answer under paragraphs of reasoning, caveats, and preamble. Instead of scrolling past three paragraphs of “let me think about this” to find the one line you needed, the output is restructured to put the answer first. It’s trending because it names a problem almost everyone using Claude Code or similar tools has run into but rarely fixes.

Good for: developers who use AI coding agents daily and are tired of hunting through verbose responses for the actual command, code snippet, or decision.

2. bilawalsidhu/gods-eye-view

github.com/bilawalsidhu/gods-eye-view — JavaScript, 35,846 stars

A browser-based 3D globe that renders real, publicly available satellite and geospatial data — flight tracking, imagery, and other live feeds — layered on a photorealistic Earth you can pan and zoom around, like a simplified open source version of a spy satellite console. It’s built on WebGL (the browser standard for GPU-accelerated 3D graphics) and open geospatial intelligence (OSINT) data sources.

Good for: developers interested in geospatial visualization or OSINT tooling, and anyone looking for a reference project for building slick 3D data dashboards in the browser.

3. alibaba/open-code-review

github.com/alibaba/open-code-review — Go, 31,744 stars

An automated code review tool that combines traditional static analysis (deterministic rule-checking, the kind that reliably flags things like null pointer exceptions or SQL injection) with an LLM agent for more nuanced review, leaving precise line-level comments on your pull requests much like a human reviewer would. It works with both OpenAI and Anthropic models and comes out of Alibaba’s own internal tooling at scale.

Good for: engineering teams that want an automated first-pass reviewer in their CI pipeline, especially ones concerned about security-relevant bug classes like thread-safety issues or injection vulnerabilities.

4. affaan-m/ECC

github.com/affaan-m/ECC — JavaScript, 260,252 stars

A performance-optimization layer for AI coding agent harnesses — it adds persistent memory, safety guardrails, and better default behaviors (“instincts”) on top of tools like Claude Code, Codex, Opencode, and Cursor, so agents behave more consistently across sessions instead of starting from scratch every time. Its star count this week reflects just how much energy is currently going into making coding agents more reliable rather than just more capable.

Good for: teams standardizing on multiple AI coding agents who want shared memory and behavior across tools instead of re-configuring each one separately.

5. openai/plugins

github.com/openai/plugins — JavaScript, 6,856 stars

OpenAI’s own reference repository for building plugins — small extensions that let ChatGPT and other agent-style products call out to external tools and APIs. It’s a lighter-weight, more official entry on this list, but official first-party repos from major AI labs reliably draw renewed attention whenever developer interest in agent tooling and extensibility spikes, as it has this week.

Good for: developers building third-party integrations for OpenAI’s agent products, or anyone comparing plugin architectures across different AI providers.

6. mksglu/context-mode

github.com/mksglu/context-mode — TypeScript, 23,227 stars

Middleware that sits between your AI coding agent and its tools, aggressively trimming noisy tool output (the project claims a 98% reduction) before it eats into the agent’s context window — the limited amount of text an LLM can “see” at once. It also persists memory across sessions and routes consistently across 17 different agent platforms using MCP (Model Context Protocol, an emerging standard for connecting agents to tools) and hooks.

Good for: anyone running long or complex agent sessions in Claude Code, Codex, or Cursor who keeps hitting context limits or losing continuity between sessions.

7. Tencent/WeKnora

github.com/Tencent/WeKnora — Go, 25,258 stars

A self-hostable platform that turns your raw documents into a queryable knowledge base using RAG (retrieval-augmented generation — a technique where an LLM looks up relevant document snippets before answering, instead of relying purely on what it memorized during training). On top of that it adds an autonomous reasoning agent and a self-maintaining internal wiki, covering the full pipeline rather than being just a vector-search library you have to wire up yourself.

Good for: teams building internal knowledge bases, support bots, or documentation search who want a batteries-included, self-hosted RAG stack instead of stitching together five separate open source projects.

8. obra/superpowers

github.com/obra/superpowers — Shell, 287,613 stars

The highest star count on this week’s list. It’s a collection of reusable “skills” and a broader development methodology designed to make AI coding agents follow better software engineering practices — planning before writing code, delegating isolated pieces of work to subagents, and generally acting less like an eager intern and more like a disciplined engineer. It’s become one of the most widely adopted frameworks for structuring how agents like Claude Code approach real engineering tasks.

Good for: teams who’ve noticed their AI agents “winging it” on planning, testing, or task breakdown and want a proven methodology to bolt on instead of writing their own from scratch.

9. DietrichGebert/ponytail

github.com/DietrichGebert/ponytail — JavaScript, 140,336 stars

A set of rules and prompts built around YAGNI (“you aren’t gonna need it” — a principle that warns against building for hypothetical future requirements) that push AI coding agents toward minimal, pragmatic solutions instead of over-engineered ones. The project’s own tagline sums up the appeal: it tries to make your agent think like “the laziest senior dev in the room,” favoring the code you never had to write.

Good for: anyone who’s watched a coding agent turn a five-line bug fix into a 200-line “framework” complete with unnecessary abstractions, and wants a lighter-touch default instead.

10. max-sixty/worktrunk

github.com/max-sixty/worktrunk — Rust, 7,922 stars

A command-line tool purpose-built for managing git worktrees — separate working directories checked out from the same repository — specifically for parallel AI agent workflows, where you might have several agents working on different branches or tasks at once. As more developers run multiple agent sessions side by side, manually juggling worktrees has become an actual bottleneck, and this fills that gap directly.

Good for: developers running multiple parallel Claude Code or Codex sessions who are tired of manually creating and cleaning up git worktrees by hand.

Put it all together and the throughline this week is clear: developers aren’t just building AI agents anymore, they’re building an entire layer of tooling around making those agents behave — trimming their output, giving them memory, keeping them honest about scope, and making it easier to run several of them at once. If you’re using an AI coding agent regularly, there’s a decent chance at least one of these ten solves a problem you’ve been quietly living with.