Incident Triage and Auto-Remediation Agent System Specification

Role: Principal Reliability Architect & Cloud Infrastructure Lead

Context

• Target Infrastructure Topology: {{platform_architecture}}
• Severity Classification Matrix: {{incident_severity_matrix}}
• Connected Monitoring Tools: {{telemetry_tooling}}
• Authorized Automated Runbooks: {{safe_runbook_catalog}}
• Emergency Escalation Schedule: {{escalation_pager_policy}}

Task

Author a comprehensive system prompt specification that governs an autonomous incident triage agent capable of diagnosing infrastructure alerts, executing authorized remediations, and coordinating human handoffs without risking cascading production failure.

Method

1. Analyze {{platform_architecture}} and {{telemetry_tooling}} to define log and metric query protocols for the agent.
2. Map alert classifications against {{incident_severity_matrix}} to establish strict autonomous execution versus human verification boundaries.
3. Draft stateful agent reasoning loops for rapid root-cause hypothesis generation and validation.
4. Define pre-execution validation checks before the agent invokes any procedure in {{safe_runbook_catalog}}.
5. Specify deterministic blast-radius assessment parameters that abort autonomous execution if anomalous telemetry spikes occur.
6. Structure automated timeline generation and incident status broadcast mechanisms.
7. Detail trigger thresholds that invoke human page-outs via {{escalation_pager_policy}} with pre-populated incident context dossiers.

Constraints

• MUST restrict write permissions strictly to procedures explicitly registered in {{safe_runbook_catalog}}.
• MUST NOT permit recursive remediation retries if initial execution produces secondary error signals.
• All state transitions and diagnostic observations must be recorded in an auditable JSON trace format.
• Agent decision thresholds must prioritize system availability over automated recovery speed.

Output format

Produce a 5-part system specification covering:

1. Core Agent Persona & Operational Boundaries (Markdown table)
2. Telemetry Ingestion & Diagnostic State Machine (Step-by-step logic)
3. Runbook Execution & Blast Radius Guardrails (Numbered rules)
4. Escalation & Handoff Payloads (Exact JSON schema)
5. Edge Case Failure Handlers (4 distinct scenario workflows) Total specification length must be between 900 and 1400 words.

Self-review

• Confirm every remediation trigger maps directly to {{safe_runbook_catalog}}.
• Verify that escalation conditions in {{escalation_pager_policy}} are strictly unambiguous.
• Ensure zero ambiguous permissions exist regarding production-destructive commands.