Interface representing an agent's runtime behavior and interaction methods. Defines how the agent processes inputs, commits messages, and manages its lifecycle.
run: (input: string) => Promise<string>
Runs the agent statelessly without modifying chat history. Useful for one-off computations or previews.
execute: (input: string, mode: ExecutionMode) => Promise<void>
Executes the agent with the given input, potentially updating history based on mode.
waitForOutput: () => Promise<string>
Waits for and retrieves the agent's output after execution.
commitToolOutput: (toolId: string, content: string) => Promise<void>
Commits tool output to the agent's history or state.
commitSystemMessage: (message: string) => Promise<void>
Commits a system message to the agent's history or state.
commitUserMessage: (message: string, mode: ExecutionMode) => Promise<void>
Commits a user message to the agent's history without triggering a response.
commitAssistantMessage: (message: string) => Promise<void>
Commits an assistant message to the agent's history without triggering a response.
commitFlush: () => Promise<void>
Clears the agent's history, resetting it to an initial state.
commitStopTools: () => Promise<void>
Prevents the next tool in the execution sequence from running and stops further tool calls.
commitAgentChange: () => Promise<void>
Unlocks the execution queue and signals an agent change, stopping subsequent tool executions.