Verification

This is the scope of checked-in verification, not proof that every snippet works in every application. Both trees have structure, frontmatter, link and packaging validation. "Selected" means only the named templates or examples are compiled; "No" means there is no dedicated executable coverage. CI results determine whether these checks pass for a particular commit.

Java fixtures target Boot 3.5.16 / Boot 4.1.0 and JDK 17 / 21. AI audit fixtures additionally pin Spring AI 1.0.0 / 2.0.0 respectively. Database CI uses PostgreSQL 16.

Skill	Shipped Java compiled	Runtime verification
agent-qa-browser-regression	No Java shipped	Response-level cases and illustrative plans; no live browser/provider execution
ai-observability	Advisor and metrics, both trees	Async saturation, shutdown, writes and failure isolation
api-versioning	No	No
configuration-properties	ClientProperties, both trees	Valid and invalid startup configuration
container-native-deployment	No	No
domain-driven-design	No	No
event-driven-messaging	No	No
flyway-migrations	SQL, not Java	Shipped migration examples, concurrent index and documented baseline setting
hateoas	No	No
hexagonal-architecture	No	No
http-interface-clients	CatalogHttpClient, both trees	Real stub HTTP server: transient/permanent status classification and read timeout
idempotency-patterns	SQL, not Java	Claim schema, contention, tenant isolation and rollback; not a payment workflow
java-supply-chain-security	Python helper, not Java	Offline CycloneDX inventory, malformed input, size/depth limits and CLI; no scanner/signature verification
layered-architecture	CreateOrderRequest, both trees	Bean Validation and HTTP validation; not the full architecture
mcp-server	Boot 4 OrderMcpTools	Compilation only; no protocol session tests
multi-module-maven	No	No
multi-tenancy	Shared-schema templates, both trees	PostgreSQL reads/writes, cache-key isolation, filter cleanup and pooled task propagation
null-safety	No	No
oauth2-resource-server	Config, extractor and identity, both trees	Signed-token issuer/audience/expiry rejection, missing scopes and email-independent account mapping
openapi-first	No	No
problem-details-rfc9457	Handler, both trees	Selected HTTP error contracts
production-observability	No	No
resilience-retry	CatalogRetryService, both trees	Real proxies: recovery, exhaustion, permanent errors and interruption
rest-api-conventions	Envelope, page DTO and handler, both trees	Pagination, HTTP errors and fixture create/retrieve
security-audit-logging	SecurityAuditRecord, both trees	Typed fields, identity requirements and correlation validation; no sink/outbox integration
spring-ai-integration	No	No provider or conversation tests
spring-ai-security	DocumentReadTool, both trees	Scoped in-memory reads, argument validation, concurrent call budget and output limits; no model/adapter tests
spring-batch	No	No
spring-boot-migration	No	No generated application migration tests
spring-cloud-gateway	No	No
spring-data-jpa	No	Fixture entity checks mappings, versioning and rollback; not every shipped snippet
spring-data-redis	Config, coalescing manager and optional lookup, both trees	Redis 7.4: concurrent loads, TTL, serialization, transaction callbacks; injected outage failures and real recovered reads
spring-incident-response	Python helper, not Java	Local hashes, changed/non-regular file rejection and CLI; no live collection or provenance proof
spring-modulith	No	No
spring-security-jwt	Templates, both trees	Token/account rejection, authorization and real HTTP authentication
spring-security-review	DocumentAccess, both trees	Identity/scope rejection and scoped in-memory lookup; no HTTP chain or real SQL for this template
testing-pyramid	No inline examples	Equivalent HTTP workflow fixture; application fixtures and Testcontainers snippets not compiled
transactional-patterns	No	Fixture rollback only; no full skill-specific suite
webflux-reactive-patterns	No	No

Agent evidence

The published smoke review records one historical Codex first-party-JWT response on an uncommitted tree with the configured default model. It is not a current-commit regression run. No Claude result is published.

Before claiming validated behavior for both agents, run all configured scenarios from a clean commit, record the model identifier, CLI version and commit, and review the rubric for each response. Separately test actual skill installation and automatic discovery: supplying a skill explicitly in a prompt does not demonstrate discovery. Keep transcripts free of credentials and private code.

Next coverage priorities: full client deadline/cancellation under load, Redis failover and cluster behavior, reactive tenant propagation, then executable slice/Testcontainers examples. These fixtures do not claim distributed-lock correctness after lease expiry, database RLS coverage, production identity-provider integration or autonomous agent benchmark results.