Handle Agent Events Skill

Process steps in order. Do not skip ahead.

Step 1 — Add the Dependency

implementation("ai.koog:agents-features-event-handler:1.0.0")

The umbrella koog-agents does not include the event handler — add it explicitly.

Proceed immediately to Step 2.

Step 2 — Install Inside the Trailing Lambda

Write the modified agent construction and the dependency to disk with explicit Path: labels (same convention as scaffold-agent):

• Path: src/main/kotlin/com/example/Main.kt — modified agent construction (or whichever file contains the AIAgent(...) call)
• Path: build.gradle.kts — appended dependency line

Create files if they don't exist. Do not respond with prose only.

import ai.koog.agents.features.eventhandler.handleEvents

val agent = AIAgent(
    promptExecutor = ...,
    llmModel = ...,
    systemPrompt = "...",
) {
    handleEvents {
        onToolCallStarting { ctx ->
            println("→ Tool '${ctx.toolName}' called with args: ${ctx.toolArgs}")
        }
        onToolCallFinished { ctx ->
            println("← Tool '${ctx.toolName}' returned: ${ctx.result}")
        }
        onLLMRequestStarting { ctx ->
            println("LLM ▶ ${ctx.model}")
        }
        onLLMRequestFinished { ctx ->
            println("LLM ◀ ${ctx.tokenUsage}")
        }
        onAgentFinished { ctx ->
            println("✓ Agent finished after ${ctx.iterations} iterations")
        }
        onAgentError { ctx ->
            System.err.println("✗ Agent error: ${ctx.error}")
        }
    }
}

Each callback receives a typed context object with what the event carried. Don't allocate inside the hot path — for high-throughput agents, an event handler that synchronously logs every tool call to stdout will dominate the run.

Proceed immediately to Step 3.

Step 3 — Don't Conflate With OpenTelemetry

Event handlers and OpenTelemetry coexist; they don't replace each other:

• Event handlers are for human-readable per-step output. Stdout logging during development; conference-demo visualization (planner step count ticking up live on a side panel); ad-hoc push to a non-OTel sink (Slack, a webhook, a custom database)
• OpenTelemetry is for production signal. Aggregated metrics (gen_ai.client.token.usage, gen_ai.client.tool.count), distributed traces, dashboards

If the user wants production observability, install both — the event handler for the development surface, OpenTelemetry for the production signal. See add-observability for the OTel install.

If the user only wants demo-time visualization, the event handler alone is enough — no OTel collector to stand up.

Finish here.