agent-health-monitoring

Monitor agent vital signs and detect anomalies in production AI fleets.

364|53|Updated May 9, 2026
One-click install
npx skills add https://github.com/cosmicstack-labs/mercury-agent-skills --skill agent-health-monitoring
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: agent-health-monitoring
Source: https://github.com/cosmicstack-labs/mercury-agent-skills/tree/main/categories/ai-ml/agent-health-monitoring
Command: npx skills add https://github.com/cosmicstack-labs/mercury-agent-skills --skill agent-health-monitoring

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Production multi-agent systems can fail silently—agents may become unresponsive, loop, degrade latency, or produce errors without triggering standard infrastructure alerts.

Core Features & Use Cases

  • Agent vital-sign monitoring: Tracks response rate, latency (P50/P95/P99), error rate, step count, tool success rate, token consumption, context pressure, and hallucination/quality proxies.
  • Liveness, readiness, and deep checks: Verifies the agent is alive, ready to accept work, and capable of producing valid outputs via diagnostic test tasks.
  • Anomaly detection and severity-based alerting: Detects statistically unusual behavior and routes incidents by P0–P3 severity to the appropriate channels.
  • Operational dashboards and runbooks: Provides the panel set and incident playbooks needed to investigate and recover quickly (unresponsive agents, error spikes, token budget anomalies).

Quick Start

Ask your agent to “Check agent health for all agents, detect anomalies from the last metrics window, and generate the monitoring dashboard plus the recommended incident response steps for any P0/P1 alerts.”

Frequently Asked Questions about agent-health-monitoring

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I detect silent failures in production multi-agent systems?

Detect silent failures by monitoring agent vital signs like latency, error rate, step growth, and token context pressure. This Skill applies liveness probing and statistical anomaly detection to identify unresponsive or degraded agents.

What metrics are used for AI agent monitoring and anomaly detection?

AI agent monitoring uses metrics like P50/P95/P99 latency, error rate, tool success rate, step count, token consumption, and context pressure. These signals are evaluated against statistical thresholds to detect anomalous behavior.

How do I set up severity-based alerting for AI agent incidents?

Severity-based alerting routes incidents by P0 through P3 severity to appropriate channels. You define alert rules using statistical thresholds for metrics like error spikes and latency degradation to trigger incident response workflows.

How can I generate operational dashboards and incident response runbooks for agent fleets?

Generate operational dashboards and incident response runbooks by analyzing agent health metrics and detected anomalies. This produces the panel set and playbooks needed to investigate and recover from unresponsive agents or token budget anomalies.

What is the difference between liveness, readiness, and deep checks for AI agents?

Liveness checks verify an agent is alive, while readiness checks confirm it can accept work. Deep checks validate the agent is capable of producing valid outputs by running diagnostic test tasks.

Do I need telemetry instrumentation to monitor agent health?

Yes, telemetry instrumentation is required to monitor agent health. It provides the operational health signals like latency metrics and token consumption needed to detect anomalies and trigger severity-based incident response workflows.