Metrics Aggregator Skill for Multi-Instance Error Rates
Normalize request counters and aggregate numerators and denominators before computing fleet-wide error rates.
中文Install in Claude Code
Use two local service instances or synthetic counters, with a single window and a shared counter definition.
- Get the complete `plugins/performance/metrics-aggregator/skills/aggregating-performance-metrics` directory from the pinned GitHub commit.
- Copy it into the target project's `.claude/skills/`. Check `.claude/skills/aggregating-performance-metrics/SKILL.md`, start Claude Code at the project root, and call `/aggregating-performance-metrics`.
Example task
Record 10 errors out of 100 requests for A and 3 errors out of 10 for B in the same window. Sum the counts before calculating the rate.What to check
The combined rate is 13 out of 110, about 11.82%. Compare it with the unweighted mean of 20%.
Original SKILL.md on GitHub · Updated
Frequently asked question
Can I just average each instance's percentage?
Weight by traffic. Sum errors and requests first so the rate represents actual requests.