Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →camunda/ci-runner-utilization
Detect CI runner underutilization and recommend downsizing for cost savings.
Passed 13 of 17 test cases with the skill, and 6 of 17 without it.
and 10 more in the full report
11 fixed6 passed either way4 still failing2 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 10 test cases the skill fixed. Same model, same prompt, with and without the skill.
A team member suggested changing a runner label to `gcp-16core-perf-default` or `aws-standard-c6i-8` to match typical cloud provider naming conventions. Explain the exact format for self-hosted runner labels in this repository and construct the label for an 8-core GCP perf runner with default durability.
| Checked | Without skill | With skill |
|---|---|---|
| gcp-perf-core-8-default | Missed | Passed |
| specifies the naming structure sequence as cloud vendor, optional tier, core count prefix, and durability level | Missed | Passed |
The instructions your model reads when the skill loads.
Analyzes CPU and memory utilization of self-hosted CI runners in the camunda/camunda repository to find overprovisioned jobs and recommend cheaper runner types.
Self-hosted runners cost money — these are Kubernetes pods on GCP or AWS billed by core-hour. Their runner_type starts with gcp- (e.g., gcp-perf-core-16-default) or aws-. More cores = higher cost. Downsizing from 16 to 8 cores roughly halves the per-job compute cost.
GitHub-hosted runners are free for public repos — jobs on ubuntu-latest / ubuntu-slim have runner_type = NULL in BigQuery. Ignore them entirely for cost optimization.
CPU is the expensive resource. Memory is proportional to cores and much cheaper per unit. Focus downsizing decisions on CPU utilization; only check memory to ensure a smaller runner won't OOM.
Perf runners cost more than standard runners — gcp-perf-core-N uses faster CPUs than gcp-core-N. Only suggest downgrading perf→standard if the job doesn't need fast CPUs (e.g., linting, static analysis, artifact uploads).
Longrunning runners cost more than default — -longrunning has higher durability guarantees and costs more. Only needed for jobs that genuinely run long or are release-critical.
Format: {cloud}-{tier?}-core-{cores}-{durability}
| Component | Values |
|---|---|
| cloud | gcp, aws |
| tier | perf (fast CPU, more expensive) or absent (standard) |
| cores | 2, 4, 8, 16 — number of vCPUs |
| durability | default (cheap, preemptible), release / longrunning (expensive, durable) |
Available self-hosted runner types can be found on https://github.com/camunda/infra-global-github-actions/blob/main/actionlint/actionlint.yaml
Downsizing follows the same family: gcp-perf-core-16-default → gcp-perf-core-8-default.
bq CLI authenticated with access to project ci-30-162810
bq query --use_legacy_sql=false 'SELECT 1'ci-30-162810.prod_ci_analytics.build_status_v2 (90-day retention)ci_url exactly. A LIKE "%camunda/camunda%" also matches camunda-optimize, camunda-docs, camunda-platform-helm and others, mixing other repos' jobs into the resultstest cases passed, vs 6 / 17 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.