SLO Implementation

/plugin marketplace add wshobson/agents; /plugin install observability-monitoring

Indicators come first and they are written as PromQL, not as intentions: successful requests over total, requests under a latency bucket, successful writes over attempted. The target is then chosen against what it costs — the difference between 99.9 and 99.99 percent is forty-three minutes a month versus four, and the skill puts that table in front of the decision. From there the error budget becomes a policy: how much of it remains determines whether risky changes ship this week or wait.