Observability & Monitoring Skill

Full-stack observability for AI agents — debug errors from logs, trace latency across services, manage alerts, respond to incidents with runbooks, and protect SLO error budgets via Datadog, Grafana, or New Relic.

Skill Standard MCP Server ADK-Rust Enterprise License

What This Skill Does

This skill orchestrates 28 observability tools into SRE workflows — from debugging a production error to declaring an incident to protecting SLO budgets.

Workflow Tool Calls What It Achieves
Debug Errors 3-4 Logs → traces → root cause
Trace Latency 3-4 Find the slow span
System Health 1-2 CPU/memory/disk overview
Alert Management 2-3 Triage + runbook + acknowledge
Incident Response 3-4 Declare + investigate + resolve
SLO Tracking 2-3 Budget remaining + forecast
Dashboards 1-2 Overview panels

Without this skill:

  • Debugging takes 30+ minutes of manual log searching
  • Alerts pile up without triage (alert fatigue)
  • Incidents declared too late (after customer impact)
  • SLO budget burned without awareness
  • No runbooks found during emergencies

With this skill:

  • Root cause found in 3-4 tool calls (logs → traces → span)
  • Alerts triaged with runbook in hand
  • Incidents declared within 5 minutes of detection
  • SLO burn rate checked before every deploy
  • Runbooks surfaced automatically for known issues

Installation

git clone https://github.com/zavora-ai/skill-observability-monitoring.git \
  ~/.skills/skills/observability-monitoring

Requirements

Required: mcp-observability (28 tools)

Backends supported:

  • Datadog (logs, metrics, traces, monitors, dashboards)
  • Grafana Cloud (Loki + Prometheus + Tempo)
  • New Relic (APM, logs, dashboards, alerts)
  • Custom API (your own monitoring stack)

Cross-MCP:

  • mcp-slack — alert escalation and incident channels
  • mcp-itsm — auto-create incident tickets
  • mcp-cicd — deploy gates based on SLO health
  • mcp-environment — rollback on degradation

Folder Structure

observability-monitoring/
├── SKILL.md                       # 200 lines — 7 workflows + decision tree
├── scripts/
│   └── alert_severity.py          # Severity classifier from error rate + latency
├── references/
│   ├── tool-sequences.md          # 28 tools across 8 categories
│   ├── cross-mcp-workflows.md     # Obs + Slack + ITSM + CI/CD + Environment
│   └── examples.md                # 3 real scenarios (debug, SLO, alerts)
├── README.md
└── LICENSE

Example

User: "Why is the payment service erroring?"

Agent behavior:

  1. Gets recent errors with stack traces
  2. Searches traces for failing requests
  3. Gets full trace — finds downstream timeout
  4. Reports root cause with next steps

Result:

Root Cause: customer-service timeout → null response → NPE at PaymentHandler:45
Started: 12 min ago | Occurrences: 23
Next: Check customer-service health, restart if overloaded

Success Criteria

Metric Target
Debug speed Root cause in 3-4 tool calls
Incident response Declare + runbook in < 5 calls
SLO awareness Always check before deploying
Alert triage Runbook found for every known alert

Scripts

alert_severity.py

Classifies alert severity from error rate and latency:

python scripts/alert_severity.py '{"error_rate": 5.2, "latency_p99": 3000}'
# → {"severity": "critical", "error_rate": 5.2, "latency_p99": 3000}

MCP Server Compatibility

Category Tools Count
Logs query, stats, errors, tail 4
Metrics query, list, health, compare 4
Traces search, get, service_map, latency 4
Alerts list, get, create, acknowledge 4
Incidents list, get, create, update 4
SLOs list, get, forecast 3
Dashboards list, get, runbook 3
Services list, get 2
Total 28

Related Skills

Contributors

James Karanja Maina

License

Apache-2.0

Part of the ADK-Rust Enterprise skills ecosystem. Built with ❤️ by Zavora AI