Everything a coding agent knows about your task has to fit in its context window. Most teams fill it without noticing. They paste the same instructions into every chat, load guidance the task doesn't need, and then wonder why the agent loses the thread.

This session shows what a model call actually sees and how to spend that space on purpose. We'll compare how instructions, skills, and agent profiles load, why a plugin or MCP server doesn't add a separate context window, and when a subagent is worth the overhead. Then we'll package skills into a plugin, share it through a team marketplace, and test that each one activates when it should and still gets the right answer.

You'll leave knowing where your agent's context goes and how to give it only what it needs.