An action the harness can execute when the model asks: read a file, run a command, search for something.
↑ ContentsDefinition
Each tool has a name and a description that the model receives in the context. When the model needs something done, it produces a tool call in its response. The harness reads that tool call, executes the action, and puts the result back in the context. Tools can be built-in (shipped with the agent, such as Read, Bash, Edit, Grep) or external (connected via MCP).
↑ Contents