Claude Agent SDK Telemetry

Implementation evidence

Claude Agent SDK Telemetry: from boundary to verified row

Use Claude Agent SDK Telemetry at a controlled application boundary, keep the event contract small, and verify a known outcome before building aggregate views.

1
Choose the outcome
Claude agent run observability
2
Define the contract
run_id, workflow, agent_name, model_alias, and prompt_version
3
Instrument the boundary
Wrap the public query or client boundary and emit one application outcome after the SDK stream reaches a terminal result; do not depend on undocumented message fields.
4
Verify the evidence
Exercise a known fixture, then inspect agent_run_completed for one correctly typed terminal row.

Before you start

Prerequisites and boundaries

@anthropic-ai/claude-agent-sdk and telemetry-sh running in a trusted server or worker process
An application-owned run ID, workflow name, and terminal outcome taxonomy
A reviewed policy for permissions, prompt content, tool inputs, tool results, and file paths

Delivery setup

Install and initialize server-side

Import telemetry-sh in server-only code and initialize it once with process.env.TELEMETRY_API_KEY. Keep ingestion credentials out of browser bundles, client-visible environment variables, source control, logs, and exception messages.

claude-agent-sdk-install

npm installation

bash

npm install telemetry-sh

1Prepare one reusable server-side delivery client with bounded network behavior.
2Add the outcome event at the success, failure, retry, or timeout boundary.
3Send controlled fixtures and inspect the stored rows before enabling an alert.

Snippet

Start with one structured event

Add this shape where the workflow completes, fails, or retries. Then build the dashboard from real fields.

claude-agent-sdk

Claude Agent SDK Telemetry event

javascript

import { query } from "@anthropic-ai/claude-agent-sdk";

async function runSupportAgent(prompt, runId) {
  const startedAt = performance.now();
  let status = "success";
  let turnCount = 0;
  let toolCallCount = 0;

  try {
    for await (const message of query({
      prompt,
      options: { maxTurns: 8 },
    })) {
      turnCount += message.type === "assistant" ? 1 : 0;
      toolCallCount += countToolCalls(message);
    }
  } catch (error) {
    status = "failed";
    throw error;
  } finally {
    await telemetry.log("agent_run_completed", {
      run_id: runId,
      workflow: "support_resolution",
      agent_name: "support_agent",
      status,
      turn_count: turnCount,
      tool_call_count: toolCallCount,
      duration_ms: Math.round(performance.now() - startedAt),
      release: process.env.APP_RELEASE,
    });
  }
}

Event contract

run_id, workflow, agent_name, model_alias, and prompt_version

status, duration_ms, turn_count, tool_call_count, and estimated_cost_usd

permission_denial_count, max_turns_reached, reviewer_outcome, release, and environment

Implementation checkpoints

Checkpoint 1

Wrap the public query or client boundary and emit one application outcome after the SDK stream reaches a terminal result; do not depend on undocumented message fields.

Checkpoint 2

Keep Anthropic's SDK observability or traces for step-level diagnostics and send only approved aggregate outcome fields to Telemetry.

Checkpoint 3

Exercise permission denial, tool failure, maximum-turn, cancellation, and process-shutdown paths before alerting on completion rate.

Verification

Prove the event arrived

Run this after exercising known success and failure cases. Replace the fallback table name if your final event contract differs from the snippet.

claude-agent-sdk-verification

Claude Agent SDK Telemetry verification query

sql

SELECT *
FROM agent_run_completed
ORDER BY timestamp_utc DESC
LIMIT 20;

Confirm one terminal row per logical outcome, with the expected status, identifiers, units, and UTC time.

Inspect the inferred schema and verify that retries do not change field types or generate a new logical event ID.

Search the stored fields for credentials, raw payloads, prompts, private content, and unbounded error messages.

Exercise a provider timeout, ingestion rejection, and process shutdown before treating the dashboard as complete.

Implementation references

Review the event contract, data-safety guidance, and upstream primary documentation before enabling a new production path.

Canonical wide events AI agent quality evaluation Agent task success SQL Claude Agent SDK overview Claude Agent SDK cost and usage

Production boundary

Keep the outcome event small and recoverable

This pattern provides

A bounded, SQL-ready outcome beside the upstream workflow.
Stable fields for dashboards, alerts, and cross-event correlation.
A fixture-driven path for validating success, failure, retry, and timeout behavior.

This pattern does not provide

An OTLP exporter, automatic collection pipeline, or replacement for detailed traces and diagnostic logs.
Exactly-once delivery merely because the payload contains an event ID.
Permission to collect raw provider payloads, user content, credentials, or regulated data.

Delivery and idempotency Batching and shutdown Troubleshoot ingestion

Event schema starting points

Review the row grain, emit boundary, required types, privacy classes, example payload, and validation checklist before adapting a query or snippet to production.

AI operations10 fields