Agent-Skills-for-Context-Engineering

Why we included this project

Context engineering is the discipline of managing everything that enters the model's attention window, and this collection of skills is a good place to learn it. Instead of a single tool, it's a structured set of skills that walk through the anatomy of context, common failure modes like lost-in-the-middle and context poisoning, and how to design agent operating loops that keep high-signal tokens in front of the model. The material is organized so you can jump straight to the skill you need when debugging a specific degradation pattern, and it's platform-agnostic, so the lessons apply whether you're working with Claude Code, Cursor, or your own harness. It's also been cited in academic surveys on agent harness engineering, a decent signal that the material is substantive rather than a list of tips. If you're optimizing long-running agents or building evaluation frameworks, the skills here apply directly.