1. reverse-skill by zhaoxuya520 - Reverse Engineering / Authorized Penetration Testing / Security Research Skill Router Pack
Language: PowerShell · Stars: 21821 · +9635 this week · View on GitHub →
What it is. reverse-skill is an AI-powered skill router pack that directs AI coding agents to the appropriate methodology, tools, and repeatable workflows for reverse engineering, penetration testing, and security research tasks across APK, binary, JS, CTF, and pentest scenarios.
Why it's trending. It gained over 10,400 stars this week by addressing a real gap—AI agents often do not know whether to use jadx, apktool, Frida, IDA, or BurpSuite for a given task—and this pack provides structured routing, on-demand toolchain bootstrapping, and a self-evolving knowledge base to make security workflows repeatable instead of ad hoc.
Who should care. Security researchers, penetration testers, reverse engineers, and CTF competitors who use AI coding clients such as Claude Code, Cursor, or Cline and want consistent, tool-aware analysis pipelines for diverse targets.
2. TencentDB-Agent-Memory by TencentCloud - TencentDB Agent Memory is a team-level memory hub for AI Agents
Language: TypeScript · Stars: 18400 · +8046 this week · View on GitHub →
What it is. TencentDB Agent Memory is a team-level memory hub for AI agents that converts conversations, documents, and code into four reusable memory assets—Chat Memory, Skill, LLM-Wiki, and Code-Graph—that are governed, shared, and equipped across agents and frameworks.
Why it's trending. The project gained 7,501 stars this week as its Team Memory Beta evolves quickly, offering one-command deployment of memory services that let agent teams accumulate, circulate, and reuse experience across sessions instead of starting from scratch each time.
Who should care. Developers and small teams leveraging AI agents who want to reduce repetitive context-setting and let accumulated knowledge, skills, and code understanding persist and transfer across agent sessions, frameworks, and team members.
3. airllm by lyogavin - AirLLM 70B inference with single 4GB GPU
Language: Jupyter Notebook · Stars: 30214 · +5711 this week · View on GitHub →
What it is. AirLLM is an open-source large language model inference tool that drastically cuts GPU memory requirements to run massive models on low-resource hardware, eliminating the need for quantization, distillation, or pruning.
Why it's trending. The project recently added support for Kimi K3 (2.8T), the largest open-source model released to date, which runs on under 4GB of VRAM via per-expert sparse mixture-of-experts streaming, following v3.0 updates that added FP8 support and compatibility with major recent models including DeepSeek-V3 and Qwen3.
Who should care. Developers, researchers, and small teams with limited access to high-VRAM GPUs who want to run large open-source language models locally without expensive enterprise hardware.