What problem does it solve? Unit and in-memory integration tests cannot catch failures that only appear over real sockets, real wire protocols, and real infrastructure — such as timeout misconfiguration, partial writes, packaging bugs in published SDKs, and non-idempotent retries. This Skill defines a production-parity e2e testing tier that runs every dependency as the real engine locally and injects network faults to verify behavior under failure. ## Core Features & Use Cases - Production-parity stack: Real PostgreSQL (dropped and recreated per run), disposable Valkey, real brokers (Redpanda, NATS, RabbitMQ), and local blockchain networks (Sui localnet, Anvil) instead of in-memory shims or client mocks. - Mandatory network-fault injection: Toxiproxy-based scenarios covering latency, black holes, connection resets, bandwidth throttling, flapping, and recovery, with assertions on deadlines, idempotency, and clean recovery. - SDK verification: Generates standalone smoke programs that consume packed SDK tarballs as real users would, catching broken exports and packaging bugs that in-process tests miss. - Use Case: Before a release, run test:e2e to spin up Postgres, Valkey, and Redpanda via Docker Compose, execute critical user journeys over real sockets, then verify the service honors its configured timeouts and retries idempotently when Toxiproxy black-holes the database connection. ## Quick Start Ask the agent to set up an e2e test tier with real Postgres and Valkey containers, a Toxiproxy fault-injection harness, and a separate test:e2e command that runs outside the per-commit fast path.
