What is an agent skill? How to protect your agent's context window.
LangChain's 'skills' pattern lazily loads agent context to prevent context window overload
“With skills, information is loaded in layers, not all at once, so context stays small until it's actually useful.”
LangChain explains an architectural pattern called 'skills'—folders of task-specific context (instructions, scripts, docs) that are lazily loaded into an agent's context window only when needed, rather than upfront. This addresses a common failure mode where long-running agents degrade because their context window is overloaded. The pattern is analogous to just-in-time loading and is framed as part of a broader 'agent harness' context management toolkit.