agent-skills-for-context-engineering

Agent Skills for Context Engineering is an open collection of skills focused on context engineering and harness engineering for building production-grade AI agent systems. It addresses the challenge of managing a language model's limited attention budget through careful curation of system prompts, tool definitions, retrieved documents, message history, and tool outputs. The repository organizes skills into several categories. Foundational skills cover context fundamentals, degradation patterns like lost-in-the-middle, and compression strategies. Architectural skills address multi-agent patterns, long-horizon prompting, memory systems, tool design, filesystem-based context, and hosted agents. Operational skills include context optimization, latent briefing, evaluation frameworks, LLM-as-a-Judge techniques, harness engineering, and self-improvement loops. Additional skills cover project development methodology and BDI mental states for cognitive architecture. The skills follow a progressive disclosure design: agents load only skill names and descriptions at startup, with full content activated on demand. They are platform-agnostic and work across Claude Code, Cursor, Codex, and other hosts supporting the Agent Skills specification. Installation is available as a Claude Code plugin marketplace or via Open Plugins for Cursor and Codex. Examples demonstrate complete system designs combining multiple skills, including a digital brain personal operating system, an X-to-book multi-agent system, LLM-as-judge evaluation tools, a book SFT training pipeline, and a long-horizon prompt lab. A researcher directory provides a file-based operating system for turning external research into skill changes, with benchmark results for skill routing accuracy across frontier models.