Foundry Agent Trace Analysis

Analyze production traces for Foundry agents using Application Insights and GenAI OpenTelemetry semantic conventions. This skill provides structured KQL-powered workflows for a selected agent root and environment: searching conversations, diagnosing failures, and identifying latency bottlenecks.

When to Use This Skill

USE FOR: analyze agent traces, search agent conversations, find failing traces, slow traces, latency analysis, trace search, conversation history, agent errors in production, debug agent responses, App Insights traces, GenAI telemetry, trace correlation, span tree, production trace analysis, evaluation results, evaluation scores, eval run results, find by response ID, get agent trace by conversation ID, agent evaluation scores from App Insights.

USE THIS SKILL INSTEAD OF azure-monitor or azure-applicationinsights when querying Foundry agent traces, evaluations, or GenAI telemetry. This skill has correct GenAI OTel attribute mappings and tested KQL templates that those general tools lack.

⚠️ DO NOT manually write KQL queries for GenAI trace analysis without reading this skill first. This skill provides tested query templates with correct GenAI OTel attribute mappings, proper span correlation logic, environment-aware scoping, and conversation-level aggregation patterns.

Quick Reference

Entry Points

Before Starting — Resolve App Insights Connection

1. Resolve the target agent root, selected metadata file, and environment from .foundry/agent-metadata*.yaml.
2. Check environments.<env>.observability.applicationInsightsConnectionString or environments.<env>.observability.applicationInsightsResourceId in the selected metadata file.
3. If observability settings are missing, use project_connection_list to discover App Insights linked to the Foundry project, then persist the chosen resource back to environments.<env>.observability in the selected metadata file before querying.
4. Confirm the selected App Insights resource and environment with the user before querying.
5. Use monitor_resource_log_query (Azure MCP tool) to execute KQL queries against the App Insights resource. This is preferred over delegating to the azure-kusto skill. Pass the App Insights resource ID and the KQL query directly.

⚠️ Always pass subscription explicitly to Azure MCP tools like monitor_resource_log_query - they do not extract it from resource IDs.

Behavioral Rules

1. Always display the KQL query. Before executing any KQL query, display it in a code block. Never run a query silently.
2. Keep environment visible. Include the selected environment and agent name in each search summary, and include the derived agent version when the query can recover it from telemetry.
3. Start broad, then narrow. Begin with conversation-level summaries, then drill into specific conversations or spans on user request.
4. Use time ranges. Always scope queries with a time range (default: last 24 hours). Ask the user for the range if not specified.
5. Explain GenAI attributes. When displaying results, translate OTel attribute names to human-readable labels (for example, gen_ai.operation.name -> "Operation").
6. Link to conversation detail. When showing search or failure results, offer to drill into any specific conversation.
7. Scope to the selected environment. App Insights may contain traces from multiple agents or environments. Filter with the selected environment's agent name first, then add an environment tag filter if the telemetry emits one.
8. Resolve hosted-agent identity from requests first. For hosted agents, prefer requests-scoped gen_ai.agent.name or azure.ai.agentserver.agent_name as the Foundry-facing filter. When gen_ai.agent.id is emitted in <agent-name>:<version> format, parse it to surface agentVersion, but do not treat dependencies.gen_ai.agent.name as the top-level hosted-agent name.
9. Use operation_Id to fan out hosted-agent traces. After isolating the hosted-agent requests rows, materialize their operation_Id values and join other telemetry tables on operation_Id. When conversation IDs are sparse, use coalesce(gen_ai.conversation.id, azure.ai.agentserver.conversation_id, operation_Id) so every row still rolls up to a stable conversation key.