SKILL.md
Scale Readiness Review
Workflow
- Establish assumptions: peak req/job rate, batch/file frequency, ledger growth/day, tenant skew, provider limits, retention needs; if unknown, state conservative assumptions.
- Inspect hot paths: N+1 queries, missing indexes, unbounded list/export/reconciliation endpoints, full-table scans in jobs, large locking transactions, sync work belonging in a job, queue jobs missing retry/DLQ/backpressure, stale-truth caches.
- Review data growth: partitioning/archival, pagination contracts, tenant isolation, rebuild paths for derived balances/projections, reconciliation strategy for drift.
- Recommend: covering index/uniqueness constraint, bounded-page batching, outbox/worker for slow effects, idempotency/retry before more concurrency, metrics for queue/latency/failure/drift, a narrow load test.
- Verify: query plans, unit/integration tests for pagination/batching/idempotency, benchmarks for hot functions, load tests where a harness exists, metrics/logs for visibility.
Output Format
Finding: unbounded reconciliation query can scan all ledger rows.
Impact: batch runtime grows with total history, not daily volume.
Fix: page by posting date and account id; add an index on (...)
Verification: run query plan and reconciliation integration test.Guardrails
- Do not optimize before the path is proven relevant to scale.
- Do not cache financial truth without freshness/invalidation rules.
- Do not increase concurrency until idempotency/locking is clear.
- Do not propose a rewrite when a bounded query, index, or queue boundary solves it.