Deep Agents adds tool binding, pinning and reloading for agent skills

LangChain says its latest Deep Agents update adds three ways to manage skills, which give AI agents task-specific instructions and files. Tools can load when an agent reads their skill; apps can pin a requested skill before the next model call or reload skills in an existing thread. On newer OpenAI and Anthropic models, LangChain says bound tools can load mid-conversation without invalidating the cached prompt prefix.