Java Enterprise Development

Java Enterprise Development

Use this skill as the Java- and JVM-specific judgement layer for long-lived enterprise systems. It turns cross-cutting architecture, security, data, delivery, and operations contracts into version-aware Java decisions and evidence without replacing their owning skills.

Prompting and AI-assisted implementation

When generating a Java task prompt, include the actual JDK/toolchain and build evidence, affected flow and layer, API/data contracts, transaction and security invariants, non-goals, expected files, and acceptance commands. Prefer a staged handoff: inspect the repository and reproduce the symptom; plan the smallest change; implement within the write set; compile and run focused tests; then review compatibility, error handling, observability, and unchanged behaviour. For debugging, provide the stack trace and reproduction, ask for a root-cause hypothesis tied to code evidence, and require a targeted regression test. For refactors, state what behaviour must remain unchanged. Treat dependency, framework, JDK, and API-version suggestions as NOT ASSESSED until the local build and current primary sources support them.

Prerequisites

Load world-class-engineering, anti-ai-slop, and the delivery evidence pack. For current versions, lifecycle, licensing, compatibility, or support claims, consult docs/source-registers/java-enterprise.md and run the Digital Research source-evaluation and source-verification gates. Load the finance engine in addition when implementing money, ledger, tax, payroll, close, or accounting rules.

Use When

• Building or reviewing Java services, APIs, workers, batch systems, libraries, modular monoliths, distributed systems, or application-server deployments.
• Working with the JVM, Spring, Jakarta EE, Hibernate, jOOQ, JDBC, Maven, Gradle, Quarkus, Micronaut, Helidon, GraalVM, WebLogic, or Oracle integration.
• Diagnosing Java CPU, memory, garbage collection, thread, connection-pool, startup, latency, or messaging incidents.
• Planning staged upgrades from older Java, Java EE, Spring Boot, Hibernate, build-tool, or application-server generations.

Do Not Use When

• The task is generic system decomposition, API design, database design, security, Kubernetes, cloud, SaaS, AI, or accounting with no Java-specific decision. Use that specialist first and return here for implementation.
• The task is Android/Kotlin application work; use the Android or Kotlin Multiplatform route unless a separate Java backend is also in scope.
• A current product or licensing claim lacks a verified source. Research it; do not fill the gap from memory or this skill.

Inputs

Workflow

1. Frame one vertical slice. Name actor, business invariant, normal path, failure path, side effects, telemetry, rollback boundary, and consequence of error.
2. Identify the actual platform. Inspect Maven/Gradle files, wrappers, toolchains, BOMs, CI, container image, framework metadata, database driver, application server, deployment descriptors, and runtime output. Record the version-awareness card below.
3. Verify lifecycle facts. Compare detected versions with the current source register. Re-open official sources when the review date or target differs.
4. Trace boundaries before code. Follow the request/event/job through transport, authentication, application rule, transaction, persistence, side effect, telemetry, and recovery. Preserve working local conventions.
5. Load only relevant depth. Select the Java reference for the decision and pair it with the owning cross-cutting skill.
6. Choose explicitly. Record an ADR for costly framework, architecture, persistence, concurrency, native-image, application-server, or vendor choices. Use measured requirements, team capability, support, and exit cost.
7. Make the smallest coherent change. Keep domain invariants visible, configuration validated, transactions explicit, concurrency bounded, and external calls failure-aware.
8. Exercise normal and failure behaviour. Run only commands supported by the repository: wrapper-based compile/test/verify tasks, static/security checks, migration checks, and a realistic smoke test. Inspect the outputs.
9. Test production shape. Where applicable exercise the packaged artefact, runtime configuration, database dialect, broker, container limits, probes, graceful termination, telemetry, and rollback or roll-forward procedure.
10. Record an evidence verdict. Use PASS, PARTIAL, FAIL, NOT ASSESSED, or NOT APPLICABLE per quality area. Never collapse missing load, failover, restore, or production evidence into a pass.

Version-awareness card

textCopy code
JDK distribution/runtime:
Java language and bytecode target:
Framework and modules:
Jakarta/Java EE level:
Persistence provider and database driver:
Build tool and wrapper:
Application server or embedded runtime:
Database and dialect:
Packaging and deployment target:
Support evidence/review date:

Quality Standards

• Correctness and data integrity outrank framework convenience. Define money, rounding, time zones, transaction isolation, idempotency, and concurrency behaviour at the business boundary.
• Prefer a supported LTS runtime for production when constraints permit, but do not upgrade or change distributions without compatibility, licensing, operational, and rollback evidence.
• Keep business rules out of controllers, listeners, entities, framework callbacks, deployment descriptors, and generated persistence plumbing.
• Treat Spring annotations, Jakarta interceptors, ORM sessions, transactions, retries, reactive pipelines, and native-image metadata as mechanisms whose boundaries and failure semantics must be understood.
• Require bounded executors, queues, result sets, request bodies, retries, timeouts, caches, batch chunks, and concurrency. Propagate cancellation or interruption according to the runtime contract.
• Verify dependencies and APIs against the pinned build. Use framework BOMs or Gradle platforms, wrappers, convergence/locking controls, and trusted repositories; arbitrary dependencies require a recorded benefit and owner.
• Make the shipped unit diagnosable: release identity, structured safe logs, metrics, traces, JVM evidence access, health semantics, and an owned runbook.
• Optimise only against a repeatable workload and before/after evidence. Do not prescribe JVM flags, reactive programming, caching, Native Image, or a new framework as folklore.

Decision Rules

Outputs

Evidence Produced

Anti-Patterns

• Adding interfaces, factories, base managers, wrappers, or mapping layers with no boundary or substitution need. Fix: keep the direct coherent design.
• Organising a large codebase only as global controllers/services/repositories. Fix: preserve existing shape or group by bounded capability with enforced dependencies.
• Treating @Transactional, retries, Kafka delivery, or database constraints as magic correctness. Fix: trace proxy/resource boundaries and test duplicate, rollback, isolation, deadlock, and partial-failure behaviour.
• Returning persistence entities as public contracts. Fix: use explicit API or event models where privacy, compatibility, or boundary ownership matters.
• Selecting microservices, WebFlux, Native Image, event sourcing, CQRS, or cache because it sounds modern. Fix: write the requirement and measure the trade-off.
• Catching Exception without containment and recovery, swallowing interrupts, logging the same failure at every layer, or exposing stack traces. Fix: translate once, preserve cause/correlation, and assign logging ownership.
• Creating raw threads, unbounded pools/queues, ad hoc retries, or parallelStream on server request paths. Fix: use managed, bounded concurrency and load tests.
• Using binary floating point for money or relying on implicit time zones and locale. Fix: define decimal scale/rounding, currency, java.time type, zone, locale, and database mapping.
• Setting heap equal to container memory or copying -XX flags from another service. Fix: budget heap and native consumers, collect evidence, change one supported control, and retain or revert.
• Claiming a build, test, benchmark, failover, or release passed without tool output. Fix: mark it NOT ASSESSED and name the exact missing check.

Read Next

• system-architecture-design, distributed-systems-patterns, and microservices-architecture for boundaries, ADRs, consistency, and topology.
• api-design-first, database-design-engineering, and the database-specific skill for contracts, access patterns, schema, SQL, and database operations.
• vibe-security-skill and web-app-security-audit for threat modelling, authorisation, supply-chain, and release security.
• advanced-testing-strategy, observability-monitoring, reliability-engineering, and deployment-release-engineering for evidence, SLOs, incidents, rollout, and recovery.
• kubernetes-platform, cloud-architecture, relevant SaaS skills, and AI skills when Java implements those specialist decisions.

References

• references/java-language-runtime-concurrency.md
• references/architecture-framework-selection.md
• references/spring-production.md
• references/jakarta-runtimes-cloud-native.md
• references/persistence-transactions-oracle.md
• references/api-messaging-distributed-systems.md
• references/security-testing-quality.md
• references/build-supply-chain-release.md
• references/performance-observability-operations.md
• references/containers-kubernetes-cloud.md
• references/oracle-platforms.md
• references/legacy-modernisation.md
• references/ai-in-enterprise-java.md
• references/source-study-notes.md
• examples/enterprise-java-reference.md
• examples/red-team-scenarios.md

Capability Contract

Read and search are required. Editing and execution follow the task authority. Network access is required for stale lifecycle/API claims. Delegation is optional and limited to independent scopes. Production mutation, deployment, data change, or irreversible migration requires explicit authority and recovery proof.

Degraded Mode

Without source, build, runtime, database, load, security, or production access, return the narrowest evidence-backed plan or review. List exact commands and artefacts still needed, and mark each affected quality area NOT ASSESSED.