| name | incident-triage |
| description | Universal first-response triage for any production incident. Classifies domain, severity, and routes to specialist agents. |
Incident Triage Skill
Step 1: Extract Signal
From the alert payload, extract:
- Service name and namespace
- Alert type (what triggered it)
- Timestamp and duration
- Any error messages in the alert body
Step 2: Classify Domain
Match alert keywords against skills/_meta/index.yaml to determine domain.
If multiple domains match, select the one with the most keyword hits.
If no domain matches, classify as unknown and load escalation skill.
Step 3: Assess Severity
- P1 (Critical): Service fully down, customer-facing impact
- P2 (High): Degraded performance, partial impact
- P3 (Medium): Internal service issues, no customer impact
- P4 (Low): Warnings, potential future issues
Step 4: Route
Based on domain classification:
- Select which specialist agents to activate
- Load the domain-specific SKILL.md for each agent
- Fan out to parallel investigation
Output Format
{
"domain": "oom|traffic_surge|deadlock|config_drift|dependency_failure|cascading_failure|unknown",
"severity": "P1|P2|P3|P4",
"agents_to_activate": ["log_analyst", "metrics_analyst", "k8s_inspector"],
"skill_path": "skills/domains/{domain}/",
"initial_hypotheses": ["hypothesis 1", "hypothesis 2"]
}