Today, when a tool call fails, telemetry records only tool_status (SUCCEEDED/FAILED/SOFT_FAIL/ABORTED) and tool_name. This is insufficient to diagnose why call-actor or any other tool fails. We need coarse failure classification across ALL tool execution paths without exposing any PII, user input, or raw error messages.
Today, when a tool call fails, telemetry records only
tool_status(SUCCEEDED/FAILED/SOFT_FAIL/ABORTED) andtool_name. This is insufficient to diagnose whycall-actoror any other tool fails. We need coarse failure classification across ALL tool execution paths without exposing any PII, user input, or raw error messages.