agent-runtime-hooks

Agent runtime hooks intercept lifecycle events across step execution, tool calls, and errors to provide tracing. You register hooks via execAgent to log each step outcome and monitor behavior without rewriting the agent core.

Yes, tool call mocking uses the beforeToolCall hook event to return deterministic mock results. This enables local testing and evaluation harnesses by intercepting tool boundaries and injecting controlled responses via event.mock().

Runtime hooks intercept human intervention lifecycle events to enable human-in-the-loop pauses. You register hook handlers to pause execution at specific boundaries, allowing manual review before the agent proceeds to the next operation.

Runtime hooks intercept context compression and sub-agent call events to monitor orchestration. You register scoped, per-operation hooks that dispatch automatically during execution and clean up when the operation completes.

No, hook handlers are fire-and-forget with non-fatal error behavior. If a hook handler encounters an error, it does not block or crash the agent execution flow, ensuring the main operation continues running smoothly.

Runtime hooks are scoped per-operation. You register them via execAgent with automatic cleanup when the operation completes, preventing hook leakage across different agent runs and ensuring isolated execution contexts.