runtime-delivery

Think Through Runtime & Delivery

Overview

Production guidance for the layer every feature stands on. Each reference paper captures the unglamorous failures: configs that validate nowhere and break at boot in production, pools that exhaust under bursts, shutdowns that drop in-flight work, deployments that skip migration ordering, and CI that proves nothing about the released artifact.

Core principle: The runtime is a contract with the operating environment. Startup, shutdown, configuration, connections, and deployment must behave correctly under partial failure, bursts, and mixed versions — or every feature built on them inherits the risk.

Domain Law

NO RUNTIME CHANGE WITHOUT:
1. the minimum required primary paper(s) for the mechanism selected from the context table;
2. failure behavior at boot, reload, shutdown, and deploy stated —
   not just the happy path;
3. "Existing-codebase checks" run against deployed configuration and
   pipeline reality;
4. every applicable MUST mapped to a decision, a gate, or a documented
   exception — never silently downgraded.

When to Use

Use this skill when thinking through, reviewing, changing, or verifying:

• project structure, module boundaries, bootstrap, dependency injection, init ordering;
• configuration: hierarchy, defaults, validation, reload, versioning, secret separation;
• connection management: pools, timeouts, reuse, backpressure;
• networking behavior relevant to services: keepalives, DNS, retries at the transport layer;
• load balancing and health-gated traffic;
• service discovery and dependency resolution at runtime;
• service-to-service communication patterns and identity;
• graceful startup/shutdown, signal handling, drain, resource cleanup;
• deployment safety: ordering, health gates, rollback, migration hooks;
• CI/CD pipelines, artifact promotion, environment parity;
• infrastructure configuration and environment provisioning.

When Not to Use

• Database/contract migration sequencing logic: use migration-evolution (030, 134).
• Telemetry emitters and alerting: use production-operations.
• Architecture style decisions (monolith vs services, boundaries): use system-architecture-harness.
• Retry/circuit-breaking policies for calls: use resilience-flow-control.