Agent Skills for Context Engineering
An open collection of Agent Skills that teach context engineering and harness engineering for production AI agent systems, covering how to curate a model's context window, design agent operating loops and evaluate agent behavior.
Agent Skills for Context Engineering is an open collection of Agent Skills focused on context engineering: managing everything that enters a language model's limited context window, including system prompts, tool definitions, retrieved documents, message history and tool outputs, rather than only crafting instructions as in prompt engineering.
The skills address predictable failure patterns that appear as context grows, such as the lost-in-the-middle phenomenon, U-shaped attention curves and attention scarcity, aiming to keep only the smallest set of high-signal tokens needed for a task. Foundational skills cover the anatomy of context, patterns of context failure, and compression strategies for long-running sessions, while architectural skills cover patterns such as multi-agent orchestration.
The repository is cited in academic papers on skill architecture and agent harness engineering, and is written in Python.