Agent Skills, Rules, Subagents: Explained!

The evolution of AI coding agents has transitioned from fragmented, complex features toward a streamlined framework centered on static and dynamic context. Rules files initially emerged to provide persistent context and mitigate hallucinations, but they often led to context bloat. Modern workflows now prioritize "skills" and "rules" as the two primary pillars for developer interaction. Rules serve as minimal, high-quality static context that evolves through feedback loops, such as tagging an agent to update configuration files. Conversely, skills represent dynamic context, packaging repeatable prompts, scripts, and third-party tools—like those found in MCP servers—into on-demand capabilities that only load when necessary. This shift reduces context window overhead while maintaining advanced functionality like OAuth and parallel processing. By compressing concepts like sub-agents, hooks, and slash commands into this binary model, developers can achieve more reliable, deterministic outcomes without the cognitive load of managing disparate agentic structures.