agent-runtime-hooks

What problem does it solve? Developers building on the agent runtime lack visibility into and control over agent execution, making it hard to observe steps, mock tool responses, handle human approvals, or track sub-agent calls without modifying core runtime code. ## Core Features & Use Cases - 16 Lifecycle Hook Types: Register hooks across 5 categories including step, tool call, human intervention, context compression, and sub-agent events via execAgent({ hooks }). - Tool Call Mocking: Use beforeToolCall with event.mock() to intercept tool execution and return simulated responses for testing. - Human Intervention Tracking: Observe approve/reject flows with beforeHumanIntervention, afterHumanIntervention, and onStopByHumanIntervention events. - Use Case: Build an evaluation harness like agent-evals that uses afterStep, onComplete, and beforeToolCall mocks to run deterministic agent tests without hitting real external tools. ## Quick Start Register an afterStep and beforeToolCall hook via execAgent to log each agent step and mock a specific tool response during a test run.