LangChain
Agent engineering platform for building LLM applications from composable parts
Small, composable agent skills from Matt Pocock's daily .agents directory
Best for: engineers who want agent skills grounded in real application development rather than in demo-friendly prompt tricks.
This repository collects the agent skills Matt Pocock uses daily, published straight from his .agents directory and released under MIT. The stated target is real engineering work rather than vibe coding.
What makes it useful: a clear design philosophy, argued explicitly. Pocock's position is that frameworks like GSD, BMAD and Spec-Kit try to help by owning the process, and in doing so take away your control and make bugs in that process hard to resolve. These skills are deliberately small, easy to adapt and composable instead — you are encouraged to hack on them and make them your own. They are also model-agnostic, so they are not tied to one provider's behaviour.
Where it falls short: small and composable also means you assemble the workflow yourself; there is no opinionated end-to-end process to follow if that is what you wanted. These are one experienced engineer's personal tools, shaped around his stack and preferences, so expect to adapt rather than adopt.
Details: Shell, MIT licensed.
Agent engineering platform for building LLM applications from composable parts
Agent harness optimisation for Claude Code, Codex, OpenCode and Cursor
Self-improving agent from Nous Research that learns across sessions
Build and train a GPT-style LLM in PyTorch, step by step