GitHub Trending Repositories — October 9, 2026
This week's GitHub trending list reveals an ecosystem where the governance layer for AI agents has become the dominant theme, but also shows projects that radically expand what's possible with visual computing, 3D reconstruction, and software portability. Here are the ten repositories that gained the most traction:
1. rea — 16.7k stars
rea is a framework that enables reverse engineering of entire applications using AI agents. Instead of relying on manual reverse engineering — static analysis, disassemblers, debuggers — it uses autonomous agents to explore application behavior, map data flows, and reconstruct functional documentation from observed execution. The idea is to treat closed binaries as living systems: interact with them, record what happens, and let agents deduce the architecture underneath.What drives rea's massive growth is the combination of two factors. First, the scarcity of documentation for corporate legacy systems is a chronic problem that no conventional tool has solved. Second, agents capable of observing behavior rather than reading code are producing practical results in scenarios where source code simply no longer exists.
2. AnyPS5 — 12.5k stars
AnyPS5 is a tool that automatically ports PS5 executables to Linux and Windows. Rather than recompiling the entire graphics engine or rewriting hardware abstraction layers, it intercepts native PlayStation 5 API calls and translates them to Vulkan and DirectX equivalents, generating a cross-platform executable from the original binary.This has implications far beyond gaming on PC. The PS5 uses an x86-64 architecture fundamentally similar to modern PCs — which makes AnyPS5 technically feasible — but the barrier has been Sony's proprietary hardware abstraction layer. By automating its circumvention, the tool opens the possibility of running any PS5 executable on consumer hardware, raising serious questions about DRM, long-term compatibility, and access to game libraries.
3. .agents — 9.8k stars
Matt Pocock's .agents repository is a collection of engineering skills and patterns for AI agents in production. The differentiator isn't in the prompts — it's in the verification gates, exit criteria, and structured workflows that prevent agents from skipping critical steps (tests, spec, security review) in favor of the shortest path.The project solves a real problem that has become visible with the popularization of coding agents: the natural tendency of agents to minimize effort in favor of speed. Well-designed skills function as checkpoints — they don't give literal code instructions, but impose a decision pattern that senior engineers apply instinctively.
4. diagram-design — 9.3k stars
diagram-design is a Claude Code extension that transforms text descriptions into editorially refined technical diagrams. Unlike generic diagram generators, it applies visual design principles — hierarchy, spacing, typography — so that outputs have publication quality for professional documentation.The rise of diagram-design reflects a practical gap: AI agents produce code and text with ease, but technical diagrams remain a visual bottleneck that requires design skills. Tools like Mermaid produce structure, not aesthetics. This project fills the space between functionality and clarity.
5. open-code-review — 8.9k stars
open-code-review, developed by Alibaba, is an AI-powered code review system that has been battle-tested across millions of internal Pull Requests. The differentiator is its focus on efficiency: the model is optimized to analyze changes in batches, prioritizing security and integrity issues before style guides.What makes open-code-review significant is the scale of its training data. Millions of internally reviewed PRs produce a corpus of code review feedback that no public dataset offers — patterns of recurring bugs, structural vulnerabilities, refactoring patterns that only appear under volume.
6. knowledge-work-plugins — 8.4k stars
Anthropic's knowledge-work-plugins are extensions for Claude that integrate knowledge work tools (research, data organization, annotation) into the text generation workflow. Instead of isolated agents, the proposal is that Claude operates as a central hub that coordinates multiple supporting tools.7. litellm — 59.8k stars
LiteLLM is an open source AI gateway that standardizes the interface for 100+ LLM providers. It translates calls to the OpenAI format, offers cost tracking, load balancing, guardrails, caching, virtual keys, and an admin dashboard — all as a centralized service.With nearly 60k stars, litellm has established itself as the most adopted abstraction layer for teams managing multiple AI providers. The project was born from the practical pain of maintaining different SDKs for OpenAI, Anthropic, Azure, Bedrock, and Vertex AI, and it solved a real infrastructure problem.
8. agent-skills — 5.5k stars
agent-skills by Addy Osmani is a framework of 24 engineering skills for coding agents, with universal CLI install that works across 70+ agents (Claude Code, Cursor, Codex, Copilot, Cline, and more). The skills cover the entire SDLC: spec-driven development, incremental implementation, testing, five-axis code review, security hardening, performance optimization, and shipping.The highlight of the project is its anti-rationalization tables — lists of excuses agents give for skipping steps, with embedded rebuttals. Instead of relying on the agent's judgment, the framework imposes checkpoints that senior engineers apply by habit.
9. artcraft — 5.1k stars
ArtCraft is an open source creative engineering suite for artists, designers, and filmmakers. The project combines 2D compositing, 3D staging, character posing, scene blocking with kit bashing, and integration with multiple image and video generation providers. The goal is to turn prompting into intentional crafting — giving precise control over composition, character identity, and camera framing.The storytelling organization also includes PhotoCraft, a clean-room Photoshop reimplementation in pure Rust, indicating a long-term strategy: building an open source creative suite as an alternative to the Adobe ecosystem.
10. lingbot-map — 17.1k stars
LingBot-Map is a feed-forward 3D foundation model for scene reconstruction from video streams, presented at ECCV 2026 (best paper award candidate). The system uses a Geometric Context Transformer that integrates anchor context, pose-reference window, and trajectory memory — three complementary context types that keep streaming state compact while preserving geometric consistency across sequences exceeding 10,000 frames.The result is ~20 FPS inference at 518×378 resolution, without needing test-time training or post-processing optimization. This makes LingBot-Map immediately applicable to SLAM, autonomous vehicles, and real-time virtual world reconstruction.
---
What this set reveals
Three trends stand out in this edition of the ranking.
First, the agent governance layer is maturing. .agents, agent-skills, and diagram-design represent different phases of the same problem: how to make AI agents follow quality engineering workflows instead of taking shortcuts. The movement spans structured skills (agent-skills) to verification gates (.agents) to qualified output (diagram-design). The question: in six months, will agent governance be as standard as linting and CI/CD are today?
Second, visual computing is opening to consumers. ArtCraft/PhotoCraft and LingBot-Map are both projects that take tools traditionally closed or computationally expensive and make them accessible — the former through clean-room Rust reimplementations, the latter through real-time feed-forward inference. The trend isn't just "AI generates images"; it's "creative tools become composable and controllable."
Third, AI agent infrastructure is becoming software infrastructure. LiteLLM, open-code-review, and rea operate at the deepest layers of development: communication with AI providers, code review at scale, reverse engineering of legacy systems. These aren't toys — they're infrastructure pieces that enterprises depend on today. litellm's consolidation as the de facto standard for LLM gateways suggests the next wave won't be new models, but new orchestration layers on top of them.
Sources: GitHub Trending, rea — morluto, AnyPS5 — boykopovar
✓ Independent sources cross-checked and verified before publishing
