What problem does it solve? Running AI agents as continuously hosted services introduces operational risks that single CLI sessions never face: uncontrolled restarts, missing audit trails, silent failure spikes, and unbounded costs. This Skill provides the operational controls needed to run agent workloads safely in production. ## Core Features & Use Cases - Lifecycle Management: Standardized start, pause, stop, and restart procedures for long-running agent processes across PM2, systemd, and container orchestrators. - Observability Baselines: Defines the metrics that matter for agent operations, including success rate, mean retries per task, time to recovery, cost per successful task, and failure class distribution. - Safety Controls: Enforces least-privilege credentials, environment-level secret injection, hard timeout and retry budgets, kill switches, and audit logs for high-risk actions. - Incident Response Pattern: A six-step failure-spike playbook covering rollout freeze, trace capture, route isolation, minimal patching, regression and security checks, and gradual resumption. - Use Case: A team deploying a customer-support agent as a systemd service uses this Skill to set retry budgets, wire audit logging for high-risk actions, and follow the incident pattern when failure rates spike after a model update. ## Quick Start Ask the agent to design an operations plan with observability metrics, safety controls, and an incident response procedure for your continuously running agent deployment.
