context-engineering-collection

Context engineering is the discipline of curating all information entering a model's context window, including system prompts, tool definitions, retrieved documents, and message history. It addresses attention limits like the lost-in-the-middle phenomenon rather than just raw token capacity.

Apply compression strategies such as structured summarization with explicit sections for files, decisions, and next steps. You can also offload tool outputs to the filesystem and use observation masking to replace verbose results with references.

Use sub-agents primarily to isolate context, not to simulate organizational roles. Supervisor, peer-to-peer, and hierarchical patterns each fit different coordination needs, but expect significantly higher token usage with multi-agent designs.

Yes, the skills are platform-agnostic and work with Claude Code, Cursor, and any agent framework supporting custom instructions or skill-like constructs. It is distributed as a Claude Code plugin marketplace and follows the Open Plugins manifest standard.

Use multi-dimensional rubrics covering factual accuracy, completeness, tool efficiency, and process quality. Combine LLM-as-a-judge for scalability with human evaluation for edge cases and end-state evaluation for agents that mutate persistent state.