Every card below was actually executed by the lab — under-the-radar repos that installed clean and did what they claim, verified in the sandbox, not guessed from the README. From 1,595 repos tested so far.
Matt Pocock's Agent Skills
A collection of specialized prompts and workflows (skills) designed to improve how AI coding agents (like Claude Code or Codex) interact with software engineering tasks.
Insight The project is a collection of markdown-based skills and a Node.js package with a clear structure and multiple distribution methods (plugin and manual install).
github.com/mattpocock/skills ↗
Git Graph for VS Code
A Visual Studio Code extension that provides a graphical representation of Git repository history.
Insight Installed cleanly on the first try; its own test suite ran — 1,266 tests passed.
github.com/mhutchie/vscode-git-graph ↗
QuantMind
QuantMind is a knowledge extraction and retrieval framework designed for quantitative finance.
Insight Installed cleanly on the first try; its own test suite ran — 334 tests passed; the demo actually ran and produced real output.
github.com/LLMQuant/quant-mind ↗
Grammar-Constrained SQL Generation
A research project implementing grammar-constrained decoding to improve the accuracy of Text-to-SQL tasks.
Insight A research project implementing grammar-constrained decoding to improve the accuracy of Text-to-SQL tasks.
github.com/aa01753039/LLM-GrammarSQL ↗
TritonGen
TritonGen is a research framework designed to evaluate how inference-time control strategies (grammar guidance, correctness feedback, and compile feedback) affect the gen.
Insight TritonGen is a research framework designed to evaluate how inference-time control strategies (grammar guidance, correctness feedback, and compile feedback) affect the generation of Triton GPU kernels .
github.com/alexeiddg/TritonGen ↗
LLM-GPU-Kernel-Generation
A framework for generating optimized Triton GPU kernels using small language models (SLMs) with grammar-constrained decoding and a three-layer validation pipeline.
Insight The project structure is complete with a clear pipeline, comprehensive test files, and a multi-layered validation logic.
github.com/casrhub/llm-gpu-kernel-generation ↗