Ops Readiness Audit

/ops-readiness-audit — Production Operations Readiness

Score /100. Verify the platform is ready for production operations: monitoring, logging, backup, alerting, deployment.

Process

1. Check Infrastructure Config

# Monitoring endpoints (broad scope — catches all submodules + infrastructure)
grep -rn "actuator\|prometheus\|metrics\|health" --include="*.yml" --include="*.yaml" \
  | grep -v node_modules | grep -v target | head -30

# Logging config (broad scope)
grep -rn "logging\.\|logback\|log4j\|winston\|pino" --include="*.yml" --include="*.xml" --include="*.ts" \
  | grep -v node_modules | grep -v target | head -20

# Backup/DR docs
ls documents/guides/*backup* documents/guides/*disaster* documents/guides/*recovery* 2>/dev/null

# Helm/k8s health probes
grep -rn "livenessProbe\|readinessProbe\|startupProbe" infrastructure/ | head -10

# Alert rules
ls infrastructure/*/alerting* infrastructure/*/alerts* 2>/dev/null

2. Primacy: bug-finding > scoring (BLOCKING)

An audit's purpose is to surface ops-operational gaps the dev team cannot trust other layers to catch. A 60/100 D score with hidden P0 gaps (no restore drill, no PII scrubber, no alert rules) is WORSE than 40/100 listing every gap honestly. Per .claude/rules/audit-skill-rubric-ops-readiness-audit.md §4.

Rules for every audit run:

1. Enumerate ALL §3 sub-checks per category. NEVER skip "obviously fine."
2. Each sub-check returns: PASS / FAIL / N/A-with-reason / ❓ UNCHECKED. No partial credit.
3. Final output starts with bug list (every FAIL surfaces) BEFORE the score.
4. Score is descriptive only; audit-level verdict = FAIL if ANY P0 sub-check FAILS.
5. If audit time-budget runs out, mark ❓ UNCHECKED — do NOT default to PASS.

3. Score 5 Categories with per-check rubric

Every category binds to a per-check pass/fail rule. Within each 20-pt category: any P0/P1 sub-check FAIL caps category total ≤ 16/20.

Per-check scoring (all 5 categories)

For each Category N:

1. Walk through every §2 sub-check in the bound rule.
2. Mark each sub-check PASS / FAIL / N/A-with-reason / ❓ UNCHECKED (no partial credit).
3. Score = 20 - (failed_P0_count * 6) - (failed_P1_count * 3) - (failed_P2_count * 1), floor 0; cap 20 if all PASS.
4. If ANY P0 sub-check fails → category total CAPPED at 16/20 AND audit-level verdict = FAIL regardless of total score.
5. Each FAIL surfaces in the audit-report bug list per §2 primacy.

Legacy scoring narrative: reference/scoring-guide.md retained for backward-compat only — the bound rule §2 IS the canonical rubric.

4. Output

Save to documents/audits/ops-readiness-audit-[date].md

Context Management

Token budget ~25-35K. Kiểm soát:

1. Grep infrastructure files — | head -20 per grep. Infrastructure files nhiều nhưng config sections lặp lại.
2. Helm values — Chỉ đọc security-relevant keys (resources, probes, securityContext), không đọc full values.yaml.
3. IaC — Chỉ check state backend config + resource count, không đọc full .tf files.
4. Doc existence check — Dùng ls thay vì cat cho backup/DR docs. Chỉ đọc nội dung nếu cần verify chi tiết.

Gotchas

• Spring Boot Actuator may be enabled but endpoints not exposed — check management.endpoints.web.exposure
• Prometheus metrics need /actuator/prometheus endpoint AND scrape config in k8s
• Message broker monitoring (RabbitMQ/Kafka): separate from app monitoring — check the broker's management plugin/exporter
• Object storage: backup strategy different from the relational DB — object storage vs relational
• If you have multiple microservices, each needs an independent health check
• IaC state: must be remote (managed object-store bucket), not local
• Multi-module scope — narrow grep on a single module may miss submodule config files; prefer broad --include="*.yml" from repo root.

Skill Contents

• reference/scoring-guide.md — Detailed rubric per category