@skelm/metrics

Prometheus-format metrics for skelm event streams.

Part of skelm.

Subscribe a MetricsCollector to any skelm EventBus and expose the resulting counters / gauges / histograms in Prometheus text format. In-process, zero managed-service dependency, zero telemetry.

Install

npm install @skelm/metrics

Quick Start

import { runPipeline, EventBus } from 'skelm'
import { MetricsCollector } from '@skelm/metrics'

const events = new EventBus()
const metrics = new MetricsCollector()
const detach = metrics.attach(events)

await runPipeline(myWorkflow, input, { events })

console.log(metrics.render())  // Prometheus exposition format
detach()

Wire render() into your HTTP server to expose /metrics:

import { createServer } from 'http'

createServer((req, res) => {
  if (req.url === '/metrics') {
    res.setHeader('content-type', 'text/plain; version=0.0.4')
    res.end(metrics.render())
  }
}).listen(9090)

What is collected

| Metric | Type | Labels | | --------------------------------- | --------- | --------------------- | | skelm_runs_started_total | counter | — | | skelm_runs_total | counter | status | | skelm_runs_in_flight | gauge | — | | skelm_steps_total | counter | kind, status | | skelm_step_duration_ms | histogram | — | | skelm_permission_denials_total | counter | — | | skelm_approvals_pending | gauge | — | | skelm_trigger_fires_total | counter | trigger |

Histogram buckets target agentic workflows, where step latency ranges from ~10 ms (deterministic code) to many minutes (LLM + tool loops): 10, 50, 100, 250, 500, 1_000, 2_500, 5_000, 10_000, 30_000, 60_000, 300_000 ms.

Stability

0.x — APIs may change between minor versions until v1.

License

MIT