Agent ops & evalsFree→ Claude / ChatGPT
Agent SLI/SLO Definer
Pick the few service-level indicators and targets that matter for your agent.
What you get: A YAML of SLIs/SLOs with measurement method, window and alert rule.
Define SLIs and SLOs for this production agent: {{agent_description}}. Users: {{users}}. Cost of a wrong answer: {{error_cost}}. Cost of an escalation to a human: {{escalation_cost}}.
Cover quality (task success, groundedness), safety (policy violations, harmful actions), reliability (error rate, availability), speed (p50/p95 latency), cost (cost per successful task) and user trust (satisfaction, override rate). For each give: the exact definition, how to measure it (automated check, judge, human sample, implicit signal), a realistic initial target with reasoning, the evaluation window, and the alert threshold with who gets paged. Identify which two indicators should be hard release gates. Output as YAML plus a short rationale.