/monitoring-audit — Observability & Incident Readiness

SkillMonitoring & ops

Observability audit — logging, error tracking, health endpoints, alerting, runbooks. Deep-dive for /launch-check monitoring.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the /monitoring-audit — Observability & Incident Readiness skill

What this skill tells your AI

The instructions your AI receives, as published by me2resh/apexyard in .claude/skills/monitoring-audit/SKILL.md and read by ahel’s review.

Deep-dive observability analysis. Checks that production issues will be detected, alerted on, and resolvable. Invoke when /launch-check's monitoring row shows WARN or FAIL.

Observability Pillars

PillarQuestionWhat to look for
LogsCan you see what happened?Structured logging, log levels, no PII in logs
MetricsCan you measure what's happening?Request counts, latency, error rates, business metrics
TracesCan you follow a request end-to-end?Distributed tracing, correlation IDs
AlertsWill you know when something breaks?Alerting rules, on-call, PagerDuty/OpsGenie

Process

Step 1: Error tracking

  • Grep for error tracking SDK (Sentry, Datadog, Bugsnag, LogRocket, Rollbar)
  • Check initialization: is the SDK configured with the correct DSN/API key (env var, not hardcoded)?
  • Check error boundaries (React ErrorBoundary, Vue errorHandler, global uncaughtException handler)
  • Check if source maps are uploaded to the error tracker for readable stack traces

Step 2: Health endpoints

  • Check for a health check route (/health, /healthz, /api/health, /_health)
  • Check what the health endpoint verifies (just "alive"? database connectivity? external service reachability?)
  • Check for a readiness probe separate from liveness (Kubernetes-style)

Step 3: Logging

  • Check for a logging library (winston, pino, bunyan, python logging, structured logging)
  • Check if logs are structured (JSON) or unstructured (console.log with string concatenation)
  • Check for appropriate log levels (error, warn, info, debug) — not everything at console.log
  • Check that sensitive data (passwords, tokens, PII) is NOT logged

Step 4: Alerting and incident response

  • Check for alerting configuration (CloudWatch Alarms, Datadog monitors, PagerDuty rules)
  • Check for a runbook (docs/runbook.md, docs/incident-response.md, RUNBOOK.md)
  • Check for a status page configuration (Statuspage.io, Instatus, or self-hosted)

Step 5: Output

MONITORING AUDIT — <project> @ <sha>

| # | Pillar | Status | Finding |
|----|--------|--------|---------|
| M1 | Error tracking | PASS | Sentry configured in lib/sentry.ts, source maps uploaded |
| M2 | Health endpoint | FAIL | No /health route found |
| M3 | Logging | WARN | Using console.log (unstructured), no log levels |
| M4 | Alerting | WARN | No alerting rules found in code or IaC |
| M5 | Runbook | FAIL | No runbook or incident response doc |

Incident readiness: NOT READY (2 fails — health endpoint and runbook are minimum viable)

Priority fixes:
  1. [ ] Add GET /health returning { status: "ok", db: "connected" }
  2. [ ] Create docs/runbook.md with deployment, rollback, and common-issue procedures
  3. [ ] Replace console.log with structured logger (pino recommended for Node.js)

Persist the run + render trend

After printing the findings table, persist via the shared audit-history lib so the monitoring trend across runs becomes legible. See docs/agdr/AgDR-0019-audit-artefact-persistence.md.

Resolve project name + score + verdict

<project-name> from apexyard.projects.yaml (or basename + /handover reminder if unregistered).

Score: score = max(0, 100 - 25*critical - 10*high - 3*medium - 1*low). Verdict by worst-severity: critical/high → fail, medium → conditional, low/none → pass. Legacy "Incident readiness" two-state: NOT READY → fail, READY → pass.

Persist + render

source "$(git rev-parse --show-toplevel)/.claude/hooks/_lib-audit-history.sh"

# Lowercase severity in the payload — the lib expects critical/high/medium/low/info.
payload=$(mktemp); cat > "$payload" <<'EOF'
{
  "schema_version": 1,
  "findings": [
    {"id": "M2", "severity": "critical", "status": "open", "summary": "No error tracking SDK detected; uncaught errors are silent"},
    {"id": "M3", "severity": "high",     "status": "open", "summary": "No /health endpoint; load balancer can't detect hung instance"},
    {"id": "M4", "severity": "critical", "status": "open", "summary": "No alerting rules; nothing pages on-call"}
  ]
}
EOF

# Body: per templates/audits/monitoring-audit.md
body=$(mktemp); cat > "$body" <<'EOF'
... (filled-in body — findings table + Priority fixes) ...
EOF

ts=$(date -u +"%Y-%m-%dT%H:%M:%SZ")
audit_run_persist "<project-name>" "monitoring-audit" "$ts" "fail" 35 "$body" < "$payload"
rm -f "$payload" "$body"

audit_render_trend "<project-name>" "monitoring-audit" 5

Opt-in commit

touch projects/<name>/audits/monitoring-audit/.audit-history-tracked

Rules

  1. Auto-PASS for projects not yet in production. Pre-launch monitoring gaps are expected — flag them as "set up before first deploy" rather than "FAIL."
  2. Check IaC too. Alerting rules might be in Terraform, CloudFormation, or CDK files rather than application code.
  3. Don't require all four pillars. Logs + error tracking + health endpoint is the minimum. Distributed tracing and detailed metrics are "nice to have" for most teams.
  4. Be specific about what the health endpoint should check — "returns 200" is table stakes; checking database connectivity is the real test.
  5. Always persist via the lib. The persist step runs regardless of opt-in commit state.
  6. Severity vocabulary in the JSON is lowercase. The lib expects critical/high/medium/low/info.

Part of ApexYard — multi-project SDLC framework for Claude Code · MIT.

Signals

GitHub stars
498
Forks
271
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
monitoring-audit
Source
github.com/me2resh/apexyard