Context Engineering Skills — контекст агентов

★ 8.0 · vibe-coding

Agent Skills for Context Engineering is an open collection of skills for managing context and building production-grade AI agent systems, and it doubles as a plugin marketplace for Claude Code. The skills teach how to curate what enters the context window, design an agent's operating loop and evaluate its behavior. Reach for it when an agent degrades over long sessions or loses information in the middle of its context; when you are assembling a multi-agent architecture with an orchestrator and workers; when an agent's tools behave unpredictably; or when you need memory ranging from simple notes to knowledge graphs. It includes the skills context-fundamentals, context-degradation, context-compression, multi-agent-patterns, memory-systems, tool-design, filesystem-context, context-optimization, evaluation and advanced-evaluation (LLM-as-a-Judge), harness-engineering, self-improvement-loops, hosted-agents, long-horizon-prompting, latent-briefing, bdi-mental-states and project-development. The agent loads only skill names and descriptions at startup and the full text when a skill is activated. Install in Claude Code with /plugin marketplace add Context Engineering Skills, then /plugin install. The principles carry over to Cursor and other platforms that support skills or custom instructions, and the examples are Python pseudocode. Not suitable when the task has nothing to do with context or the agent loop.