#### 1. AnyPS5 — PS5 Native on Any PC
boykopovar/AnyPS5 (C++, 16,000+ stars) — Convert PS5 ELF binaries to run natively on Linux and Windows, with zero emulation overhead.
AnyPS5 is not an emulator in the traditional sense. It takes executables compiled for the PS5's Zen 2 APM2 architecture and relinks them for x86-64 Intel/AMD hosts, converting APM2-specific instructions (EXTRQ, INSERTQ, MOVNTSS, MOVNTSD) at relink time — when an equivalent encoding exists, or generating jump stubs otherwise. The project implements PlayStation system prx libraries (libc.prx, libscf.prx) in native code, allowing titles to run as real OS processes.
The technical breakthrough is in the compatibility layer: instead of translating assembly instructions at runtime (as an emulator would), AnyPS5 operates at the linking level, rewriting the binary before it is loaded. This results in dramatically lower overhead — the game Dreaming Sarah runs stable at 60 FPS on modest hardware (GTX 1050 Ti, i5-7500).
The project has 679 commits and a growing game compatibility list. The SPIR-V shader recompiler enables compatibility with desktop GPU drivers, and SDL-based controller mapping simplifies input. It is the same class of reverse engineering that made RPCS3 work with PS3, but with a more architecturally clean approach.
#### 2. REA — Reverse Engineer Anything with AI Agents
morluto/rea (TypeScript, 27,500+ stars) — CLI and MCP server for making AI agents reverse engineer any software, from desktop apps to native binaries.
REA solves the reverse engineering problem for the modern agent coding workflow. Instead of a researcher learning Ghidra, Hopper, and various standalone analysis tools, REA orchestrates these backends through a unified interface of 50 tools. The agent receives commands like `open_binary`, `binary_overview`, `search_strings`, `search_procedures`, `list_names` — and REA manages installation, configuration, and routing to analysis providers.
What makes it unique is the structured investigation model: REA distinguishes observations from inferences, and results are typed (Result/ok/err) instead of raw text. This means an agent can ask "how does search work in Notes.app?" and receive structured evidence that can be reused — not just a dump of strings, but a graph of functions, imports, and cross-references.
The Hopper (proprietary analysis) and bring-your-own Ghidra adapter support gives flexibility. The Mermaid architecture diagram shows well-separated layers: domain, contracts, providers, application, server, adapters. It is a direct response to the growing need to understand code without source code — something that will become more common with closed model restrictions and increasing use of proprietary APIs.
#### 3. RAD Debugger — Epic Games' Native Debugger
EpicGames/raddebugger (C, 8,100+ stars) — Native, user-mode, multi-process, graphical debugger from Epic Games.
The RAD Debugger is not just another generic tool — it is Epic Games' response to the debugging problem in massive projects like Unreal Engine. It supports asynchronous process control, stepping, and breakpoints for all attached processes, running in lockstep with the target processes.
The big technical leap is RAD Debug Info (RDI): a custom debug information format, rather than relying exclusively on PDB or DWARF. For huge executables that break internal PDB tables (32-bit overflow), the RDI eliminates that limitation. Epic's linker generates traditional PDBs but can also create RDI natively.
The project is in alpha, but the roadmap is ambitious: native Linux support with DWARF, and full portability of the process control abstractions. If Epic open-sources all of this, we could have a GDB/Visual Studio Debugger alternative that could gain traction independent of the engine ecosystem.
#### 4. Skills for Real Engineers — Matt Pocock's Own Skills
mattpocock/skills (Shell, 281,000+ stars) — The exact skills Matt Pocock uses daily in his .agents directory.
With more than 281 thousand stars, this is likely the most starred repository on the list. Matt Pocock (creator of Zod and TypeScript training author) published the exact set of skills he uses daily — `.agent` files that define prompt templates, constraints, and workflows for his coding agents. The concept is simple: instead of writing the same prompt every day, you version-control the context, rules, and expected format.
The value here is not in any specific technical function, but in the philosophy of "skills for real engineers": documents read and modified by humans, not just consumed by machines. The idea that a `.agent` file is as important as a Makefile or `.eslintrc` is a cultural shift in the development workflow.
#### 5. claude-mem — Persistent Memory for Agents
thedotmack/claude-mem (TypeScript, 98,500+ stars) — Persistent context across agent sessions: captures what the agent does, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode, and more.
Every coding agent (Claude Code, Codex, Gemini) forgets everything when a session ends. claude-mem solves this by capturing what the agent does during a session, compressing it with AI, and injecting the relevant context back into future sessions. It works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode, and more.
The core idea is transforming the ephemeral state of agents into persistent memory — and AI compression is the trick: instead of saving raw 500MB logs, the system summarizes sessions into key concepts and decision patterns. This drastically reduces token cost for context reconstruction.
#### 6. ArtCraft — Intentional Crafting Engine
storytold/artcraft (Rust, 8,100+ stars) — An intentional crafting engine for artists, designers, and filmmakers.
ArtCraft is a procedural generation engine focused on controlled creative intent. Unlike random asset generators, ArtCraft lets creators define intentional constraints — styles, palettes, compositions — and generate variations within a defined creative space. Built in Rust for performance in real-time generation.
The repository grew over 2,000 stars in a single day — reflecting the latent pain of artists and designers who need to generate consistent assets without losing creative control.
#### 7. Diagram Design — Editorial Diagrams for Claude, Codex, Copilot
cathrynlavery/diagram-design (HTML, 46,400+ stars) — 42 types of editorial diagrams for Claude Code, Codex, GitHub Copilot, Factory Droid, and Pi. Self-contained HTML + SVG. No shadows. No Mermaid slop.
This project tackles a specific and real pain point: when LLMs generate diagrams using Mermaid, the output is often broken, illegible, or non-editable. diagram-design offers professional templates in pure HTML+SVG, ready to use — no dependencies, no runtime, no shadow DOM.
42 diagram types cover architecture, flows, ER, sequence, deployment, and more. The fact that it has 46,000+ stars shows how massive the need for professional visual documentation in AI projects has become.
#### 8. Knowledge Work Plugins — Anthropic's Plugin Ecosystem
anthropics/knowledge-work-plugins (Python, 27,600+ stars) — Open source plugin repository for Claude Cowork, focused on knowledge workers.
Anthropic opened its plugin repository for Claude Cowork — the company's integrated workspace environment. Plugins cover everything from document analysis to report generation, with focus on knowledge workflows. The project follows a modular plugin pattern, allowing anyone in the community to contribute domain-specific extensions.
#### 9. System Design Notes — Interview Prep Notes
liquidslr/system-design-notes (N/D, 24,600+ stars) — Notes for System Design Interview — An Insider's Guide by Alex Xu.
Notes from this reference book for engineers preparing technical interviews continue to trend. With 24,600 stars, it shows that system design interview preparation remains a massive priority in the tech job market — and that the open-source version of notes serves as widely accessible study material.
---
What This Set Reveals
Six of the nine repositories at the top reflect a clear pattern: code agents are becoming infrastructure. REA, claude-mem, Matt Pocock's Skills, diagram-design — all exist to solve problems in agent coding workflows. AnyPS5 and Epic Games' RAD Debugger are deep technical tools that also leverage agents for automatic analysis (REA explicitly, AnyPS5 implicitly in its porting automation).
Secondly: reverse engineering is being democratized. REA is the most emblematic tool here — no longer just for security researchers, but for any developer who wants to understand how something works without source code. Combined with AnyPS5, which reverse engineers an entire console, the movement is clear: the barrier between "what I see working" and "what I can build" is shrinking.
Finally: the coding agent ecosystem is maturing. Versionable skills, persistent memory, professional diagrams — each of these repositories solves one piece of the puzzle. Together, they show that the question is no longer "how do I use an agent" but "how do I build a reliable infrastructure around them."
Sources: GitHub - boykopovar/AnyPS5, GitHub - morluto/rea, GitHub - EpicGames/raddebugger
✓ Independent sources cross-checked and verified before publishing
