Agent framework
A library or runtime that orchestrates a language model through multi-step tasks — planning, calling tools, observing results, and looping.
Agent frameworks (LangChain, LlamaIndex, the various "agents" SDKs, and many others) provide the loop around a model: they turn a goal into a sequence of tool calls, feed outputs back in, manage memory, and decide when the task is done. Many now speak MCP as their tool layer.
Why it matters
The framework decides how much autonomy the model has — whether a tool call needs approval, how many steps it can take unattended, what happens on error. Weak defaults here turn a prompt injection into real action: the model reads a malicious instruction, the framework dutifully executes the resulting tool calls, and no human is in the loop. Most "the agent did something bad" incidents are really "the framework let it."
Exposure map
Live counts of instances showing this pattern will appear here once the exposure map is collecting data.