agent-runtime-hooks

You can trace agent execution by registering lifecycle hooks dispatched via HookDispatcher to capture beforeToolCall, afterToolCall, and afterStep events for auditing and debugging without altering core agent logic.

Yes, you can intercept tool execution and inject deterministic mocks using the beforeToolCall hook in local mode, ensuring reliable and repeatable agent testing outcomes.

Human approval workflows are integrated by registering hooks that observe pauses for approval or rejection, allowing you to intercept execution and inject manual intervention logic seamlessly.

Context compression tracking is handled by registering hooks for beforeCompact, afterCompact, and onCompactError events, which lets you monitor and manage context window reduction phases during execution.

Yes, sub-agent call visibility is supported, allowing you to trace parent operation behavior when calling sub-agents, including capturing both successful executions and error cases.

Handlers must be non-fatal async Promise<void> functions, meaning they observe and intercept events without blocking the core execution pipeline, and errors in handlers do not halt the agent runtime.