API Health Monitor Skill

Turn a wall of HTTP probe results into a one-paragraph "what is broken and why" — clustered by region, service, or status code.

The Problem

When 4 of 22 endpoints fail, the question is rarely "which 4" — it is "what do they have in common?" A human eyes the list and spots the pattern in 5 seconds. AI does the same automatically and scales as you add endpoints.

How It Works

1. A runtimeenv task hits each endpoint with curl and emits a JSON list of results
2. The api-health-monitor skill clusters failures and identifies likely root causes
3. The message task delivers the digest

Setup

1. Install this skill
2. Edit the endpoint list in the YAML below — name, region, service, URL per line
3. Schedule for whatever cadence makes sense (5 minutes for canaries, 15 minutes for production)

Example YAML Job

name: api-health-check
version: 1.0.0
description: Multi-endpoint health probe with AI clustering

notifications:
  - userNames: [intouch]
    alertOnError: true

tasks:
  - name: probe-endpoints
    tool: runtimeenv
    properties:
      scriptContent: |
        #!/bin/bash
        endpoints=(
          "checkout-api-us|us-east-1|checkout|https://api.example.com/checkout/health"
          "checkout-api-eu|eu-west-1|checkout|https://eu.api.example.com/checkout/health"
          "inventory-us|us-east-1|inventory|https://api.example.com/inv/health"
          "inventory-eu|eu-west-1|inventory|https://eu.api.example.com/inv/health"
        )
        echo "["
        first=1
        for e in "${endpoints[@]}"; do
          IFS='|' read -r name region service url <<< "$e"
          [ $first -eq 0 ] && echo ","
          first=0
          result=$(curl -s -o /dev/null -w "%{http_code} %{time_total}" --max-time 10 "$url" || echo "000 -1")
          read -r code t <<< "$result"
          status=$([ "$code" = "000" ] && echo "TIMEOUT" || echo "$code")
          ms=$(awk "BEGIN{printf \"%d\", $t*1000}")
          printf '  {"name":"%s","region":"%s","service":"%s","url":"%s","status":"%s","responseTimeMs":%s}' \
            "$name" "$region" "$service" "$url" "$status" "$ms"
        done
        echo
        echo "]"

  - name: digest
    tool: skill
    properties:
      skillName: api-health-monitor
      input: "{{probe-endpoints.output}}"

  - name: send
    tool: message
    properties:
      subject: "API health"
      body: "{{digest.answer}}"
      userNames: "intouch"

Customization

• Suppress on green — wrap the message task in a workflow tool that only fires when the digest does not begin with "all N endpoints healthy"
• Severity tiers — add a tier field per endpoint (canary, critical, internal) and prompt the skill to weight failures by tier
• Cadence — every 1m for canary, 5m for prod, hourly for low priority

Cost

• AI assistant: ~$0.001 per check with Claude Haiku — at 5-minute cadence, ~$0.25/month