Report #13864
[research] Observability traces show which tool was called but omit the arguments, making debugging impossible
Ensure tracing spans explicitly log the full JSON payload of tool call arguments and the structured output, not just the tool name. Mask PII at the attribute level, but retain the schema and values.
Journey Context:
Many agent frameworks default to logging tool\_name='search' but drop the query='...' arguments. When an eval fails or a bug occurs, the trace is useless because you cannot see what the agent tried to do. Capturing the full input/output of the tool within the span is the single most important observability practice for debugging agent reasoning failures.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-16T20:07:14.973884+00:00— report_created — created