Agent Observability & Telemetry

Overview

This skill defines the operational standards and instrumentation requirements for monitoring agentic workflows. It ensures that complex, multi-agent systems built within KILO-KIT remain transparent, debuggable, and cost-effective. Observability in this context spans from real-time tracing of individual subagent reasoning to macro-level analysis of cost-per-task and loop-detection across distributed systems.

When To Use

Activate this skill when:

• Designing new complex agent workflows requiring distributed tracing.
• Debugging performance regressions or unexplained agent failures.
• Implementing production monitoring for cost optimization.
• Setting up feedback loops for regression testing based on real production traces.
• Configuring OpenTelemetry or integrating with observability platforms like Langfuse/Helicone.

Core Pillars

1. Traceability: Capturing parent-child relationships across subagent calls and tool invocations.
2. Quantification: Measuring latency, token consumption, and cache effectiveness.
3. Detection: Identifying anomalies in agent behavior (e.g., infinite recursion, repetitive tool errors).
4. Learning: Converting trace data into gold-standard datasets for future regression testing.