← Files LLM & Agent Builder CopilotARCHIVED FILE

references/tracing-observability.md

880 Bytes · Nov 21, 2026 · 12:41 UTC

↓ Download file

See the change to this file →

# Tracing and observability
Record stable run/trace IDs, request class, versioned model/prompt/tool schemas, start/end, latency, token and cost counters when available, tool attempts, sanitized results, guardrail decisions, approvals, handoffs, checkpoint transitions and final success. Link to external audit IDs for mutations; do not store raw secrets, unnecessary personal data or untrusted full texts.

OpenAI Agents SDK includes traces for model calls, tools, handoffs and guardrails, but implementation and platform settings affect capture, retention and availability. Consult current documentation (https://openai.github.io/openai-agents-python/tracing/). Tracing alone does not substitute for a durable side-effect audit record. Define dashboards for completion rates, tool errors, retry storms, approval denials, cost and incidents. Include sampling and privacy reviews.

SHA-256: 8c0face1f1a96d526b6536a7f803cb00f3847a16dbbf04080a9f09ac73eba61d