exeris-systems

Architectural review for Exeris Kernel changes that touch module boundaries, SPI/Core/Community/Enterprise placement, dependency direction, provider wiring, or ADR-fixed structure. Use when a change adds or moves interfaces/providers, alters module layering, or risks a The-Wall breach (SPI implementation leak, Core driver leak). Not needed for pure intra-module edits with no boundary impact.

Java 26+ idiom review for Exeris Kernel. Use when a change touches concurrency, context propagation, native interop, memory APIs, or runtime data carriers — to enforce ScopedValue over ThreadLocal, StructuredTaskScope over ad-hoc async, Panama FFM (MemorySegment/Linker/SymbolLookup), Valhalla-ready immutable carriers, early construction, and removal of legacy framework idioms.

Performance-lawyer review enforcing No Waste Compute for Exeris Kernel. Use when a change touches runtime hot paths (transport, flow, memory, persistence, bootstrap, dispatch) or can affect allocations, structured concurrency, off-heap copies, or banned APIs (ThreadLocal, ExecutorService, CompletableFuture, ByteBuffer, Unsafe). Skip for docs-only or pure test/tooling changes.

Pre-PR gate for Exeris Kernel — run BEFORE opening or pushing any PR. Enforces lint-gate integrity (skip flags poison a green build's lint proof), the architecture guard test, fresh-branch-per-task discipline, and the register-ADR-before-claiming rule. Use whenever about to commit/push a branch or open a PR in the kernel repo.

Default entry-point for ANY Exeris Kernel PR/diff/branch review — start here. Triages software inflation, >5-dependency classes, and boundary/perf/contract risk, then dispatches to the focused lens skills (exeris-architect-guardrails, exeris-performance-contract, exeris-tck-first, exeris-jfr-telemetry-review, exeris-java26-panama-loom, exeris-subsystem-specialist) as the diff warrants. Trigger whenever asked to review, look at, audit, or assess a change.

Milestone release flow for Exeris Kernel — flip versions, write release notes, integrate development/X.Y.0 into main via a single release(x.y.z) PR, then open the next development line. Use when asked to prepare, cut, integrate, or finalize a milestone release, or to open the next development branch.

Targeted single-subsystem review for Exeris Kernel via one mode-driven skill — one mode per contract doc in docs/subsystems/ (bootstrap, config, crypto, events, exceptions, flow, graph, http, memory, persistence, security, telemetry, transport). Use when a change is concentrated in one subsystem and you want its contract doc loaded as the primary source of truth plus mode-specific deep checks — without spawning multiple agents. State the mode (e.g. "review this as Transport").

Run the tagged test gates (integration / continuity / stress) that the default build EXCLUDES, mapped from what a change touched. Use before claiming definition-of-done when touched code is covered by @Tag'd tests, or to reproduce a CI gate (persistence RLS, Kafka, recovery continuity, transport stress) locally.

Comparative strict-gate review for exeris-benchmarks. Use on every PR making a cross-target comparative claim — verifies 4 gate artefacts present + `claim-status.json = comparison_eligible` + `track_id` consistent.

Fairness + reproducibility review for exeris-benchmarks. Use on every committed result and every cross-target claim — enforces matched scoping, separation axes, and complete metadata.

Publication-mode scrub review for exeris-benchmarks. Use whenever `publish-report.sh` runs, when an artefact crosses to a publishable surface, or when raw `.jfr` / flamegraph appears.

Router/Planner skill for exeris-benchmarks. Produces primary agent, secondary handoffs, execution order, validation gates, and minimal next action for a benchmark task.

Router/Planner triage skill for exeris-benchmarks. Classifies task type (architecture / scenario / driver / JMH / target / verification / docs / publication), scope, severity, and recommends primary agent based on primary risk against fairness / reproducibility / strict gate / confidentiality.

TLS comparator label discipline review for exeris-benchmarks. Use on every TLS JMH benchmark / TLS row / TLS report change.

Guard Exeris cross-tier comparisons so Community vs Enterprise claims stay correctly labeled, protocol-consistent, and inference-safe.

Protect proprietary Enterprise details in benchmark outputs by reviewing traces, symbols, and publication artifacts before public reporting.

One-pass sweep of all five platform contracts for a broad or multi-module change — no-parallel-metamodel, LSP wire surface, idempotent write-back, open-core boundary, and frontend projection — returning a single consolidated verdict. Use on a large PR, a pre-merge gate, or any change whose blast radius crosses studio-backend, exeris-platform-lsp, and exeris-studio-frontend at once, when running the reviews individually would be slow or easy to forget one.

Validate a change that spans the Maven reactor and the npm frontend (cross-build). Use whenever a change touches both Java modules (exeris-platform-lsp / exeris-studio-backend) and exeris-studio-frontend, alters a contract surface the frontend consumes (an `exeris/*` LSP method or its wire shape), or bumps a cross-repo `eu.exeris:*` / `eu.exeris.tooling:*` dependency. Confirms both builds stay green and the wire contract matches on both sides before the change is called done.

Pick the right decision-document shape BEFORE drafting — Research vs RFC vs ADR — per the exeris-docs templates and the single ecosystem-wide ADR namespace. Use when asked to "write an ADR", "document this decision", "draft an RFC", or "write a research note" for the platform, especially when upstream measurement or option comparison is still missing. Enforces reserve-number-first discipline for true ADRs and steers premature ADRs back to Research/RFC.

Ensure the Studio frontend PROJECTS the canonical DomainMetadata into an ephemeral view-model rather than PERSISTING a parallel shape, and never writes `.java` sources directly. Use on any exeris-studio-frontend (Angular + embedded React) change that adds a model/store/state type, a service that holds entity/field/relation shape, or any file-write path. Complements no-parallel-metamodel-review, which guards backend/LSP rather than the frontend.

Deep, evidence-gathering review that the same `MutationOp` applied twice converges to identical on-disk state. Finds the write-path diff itself, checks writer routing, import/line/whitespace stability, and the round-trip test, then issues a verdict. Use whenever the `exeris/applyMutation` path, the `exeris-sdk-source-model` writer integration, a new mutation kind, or import/whitespace/line-number-sensitive emission changes. (For a quick inline audit with a diff in hand, the `/idempotent-writeback-check` command is the lighter path.)

Deep, evidence-gathering review of the LSP wire surface — finds the diff itself, audits namespacing, SDK-owned shapes, and transport parity, and issues a verdict. Use whenever a custom `exeris/*` method is added/removed/renamed, the wire shape of `MutationOp`/`MutationResult` changes, stdio vs WebSocket transport could diverge, or a method is consumed by `exeris-ai-bridge`. (For a quick inline audit with a diff in hand, the `/lsp-protocol-purity` command is the lighter path.)

Deep, evidence-gathering review that detects parallel-metamodel regression in `exeris-studio-backend` or `exeris-platform-lsp` — finds the diff itself, classifies each new type as projection vs persistence, and issues an APPROVE / CONDITIONAL / REJECT verdict. Use on any change that adds a record/class to those modules, adds a backend REST/HTTP endpoint, or is motivated by "store workspace entities" / "track project structure". (For a quick inline audit with a diff in hand, the `/no-parallel-metamodel-check` command is the lighter path.)

Deep, evidence-gathering review of the open-core boundary — finds the diff itself, distinguishes an extension point (open-core OK) from a concrete premium implementation (belongs in `exeris-platform-enterprise`), and issues a verdict. Use whenever a change proposes or touches a feature with a premium counterpart — multi-env promotion, RBAC, approval workflows, audit dashboards, multi-tenant/org concepts. (For a quick inline audit with a diff in hand, the `/open-core-boundary` command is the lighter path.)

Maven Central publish-readiness review for exeris-sdk. Use whenever you bump a module version, touch `<distributionManagement>` or required POM metadata (`<url>`, `<licenses>`, `<developers>`, `<scm>`, `<issueManagement>`, `<organization>`), or add a new publishable Maven module — before a release goes out.

AST + Jackson 3 wire-format contract review for exeris-sdk. Use whenever you add/change/remove an AST record under `eu.exeris.sdk.sourcemodel.ast.*`, touch Jackson config (`@JsonInclude`, serializers/deserializers), or bump `jackson-annotations` / `jackson.version` in the BOM — during editing, not only at PR review.

Field/Validation canonical scoping review for exeris-sdk. Use whenever `@Field` or `@Validation` attributes (or the `FieldMetadata` AST record) are touched, or when generator-derived NOT NULL / not-blank semantics (from `FieldMetadata.required`) are in scope.

Triage skill for exeris-sdk — use FIRST on any non-trivial SDK task, before editing annotations, AST records, UI kit, poms, or docs. Classifies task type (annotation contract / AST wire-format / UI kit / stability / docs / publish-readiness), scope, and severity, and routes to the right specialist agent.

Annotation contract review for exeris-sdk. Use whenever you add, change, or remove a public `@interface` in `exeris-sdk-annotations` (any package — root, `system`, `security`, or a newly introduced subpackage) — enforces `@Retention(SOURCE)` + `@Target`, the invariant every downstream consumer depends on.

Build-invariants review for exeris-sdk. Use whenever you touch `maven.compiler.release`, the `jacoco-maven-plugin` version, the reactor `<modules>` list, or root/parent build-plugin config — guards the JDK 26 floor, JaCoCo ≥0.8.14, and ui-kit-is-npm-only.

Coverage-gate review for exeris-sdk. Use whenever you add new code/branches to `exeris-sdk-source-model` (`ast.*`, `mutation.*`) or `exeris-sdk-ui-kit`, add or restructure a test, or hit a JaCoCo (85% BUNDLE) / Vitest (85% per-file) failure — enforces where new tests belong.

Deprecation pipeline review for exeris-sdk. Use whenever public API (annotation, AST record, attribute, public method) is removed, renamed, or moved between canonical annotations (e.g. `@Validation.X` → `@Field.X`).

Angular v22 emission discipline for exeris-codegen-ts. Use whenever authoring or modifying a TS generator (*-gen.ts) or the emitted app scaffold — keeps emitted Angular idioms on the v22 canon (Signal Forms, httpResource, dropped deprecations) and gates the phased A→B→C migration.

Decision-document shape gate for exeris-tooling. Use BEFORE drafting an ADR, RFC, or research note — picks the right shape (Research vs RFC vs ADR) for the question, enforces number reservation in the central adr-index.md, and blocks refactor-only or measurement-missing material from becoming an ADR. Invoke whenever someone asks to "write an ADR", "document this decision", or capture a design choice.

Generated-output lifecycle discipline for exeris-tooling. Use whenever a change touches how code is emitted to or pruned from src/main/generated, proposes "always regenerate" semantics, or affects the L1-committed / L2-detach story (OutputWriter, the codegen Maven plugin generate/detach mojos, the generated-output pruner, mvn clean interactions). Guards hard-constraint

After triage (see exeris-tooling-task-classifier), use to turn a classified exeris-tooling task into an execution plan — primary agent, ordered secondary handoffs, validation gates tied to the risk surface, and the minimal next action. Invoke when a build-time task spans modules or needs an explicit agent sequence and gate list before implementation starts.

Strict-mode completeness-audit review for exeris-tooling. Use whenever a generator (Java or TS) starts or stops consuming an annotation attribute or a whole annotation, when an SDK annotation surface changes, or when the INERT_ATTRIBUTES / INERT_ANNOTATIONS registries in ExerisDomainProcessor are touched. Catches stale registry entries that produce false "no effect" warnings under -Aexeris.strict.

Use FIRST, before starting any exeris-tooling build-time change, to triage it — classifies task type (processor / codegen-java / codegen-ts / e2e / docs), scope, severity, and primary risk against the pipeline contract, then recommends the primary agent. Invoke whenever a request lands in exeris-tooling and the owning module or risk surface is not yet obvious. Chain into exeris-tooling-routing-planner for the execution plan.

Codegen determinism review for exeris-tooling. Use for every PR that touches `*Generator` emission surfaces, scaffold helpers, or text-stable iteration. Catches timestamps, randoms, hash-iteration leakage, and locale-dependent formatting before they reach committed `src/main/generated/`.

Java/TS emitter parity review for exeris-tooling. Use when DomainMetadata shape changes, when a new generator surface is added on one side, or when one side starts consuming a field the other does not emit.

Detect documentation and ADR drift against runtime ownership and mode reality. Classifies drift (NO_ACTION / MINOR_DOC_UPDATE / DOCS_UPDATE_REQUIRED / ADR_IMPACT / NEW_ADR_REQUIRED), names affected docs, and recommends merge timing for the docs/ADR delta.

Prevent Spring leakage into kernel boundaries and preserve The Wall. Verifies no Spring imports/types/annotations in kernel SPI/Core touchpoints, no framework-coupled contracts in Exeris kernel abstractions, no bypass of canonical Exeris provider discovery, and no kernel API changes that assume servlet/reactive/JDBC framework runtime.

Prevent confusion between Pure Mode and Compatibility Mode in code, tests, and docs. Flags features that lack a mode label, compatibility behavior that leaks into the pure path, and docs that claim pure semantics while implementation is compatibility-driven.

Enforce module-responsibility boundaries across exeris-spring-boot-autoconfigure, exeris-spring-runtime-web/-tx/-data/-actuator. Flags web/runtime logic in autoconfigure, tx semantics in actuator, persistence ownership in web, compatibility sludge spread across modules, and runtime orchestration moved into the boot layer.

Check whether a change preserves the runtime-ownership truth (Spring = application framework, Exeris = runtime owner). Flags ownership inversion to servlet/reactive/JDBC-first runtime, split-brain lifecycle, fake host-runtime claims, and hidden fallback ownership outside Exeris.

Produce a minimal, risk-aware execution order across exeris-spring-runtime agents — choose the primary agent, ordered secondary handoffs, a 3-5 step plan, validation gates, and the single best next action. Use after task-classifier has labelled the work.

Review request-path integration overhead in Spring bridges — wrapper/DTO churn, request/response copy inflation, codec layering overhead, reflection-heavy hot paths, context propagation cost, and hidden fallback overhead to legacy runtime path. Use when a change materially affects the request path.

Classify an incoming exeris-spring-runtime task into one primary class (ARCHITECTURE / INTEGRATION_IMPLEMENTATION / VERIFICATION / DOCS_ADR / PERFORMANCE / MULTI_DOMAIN) plus mode impact (PURE_MODE / COMPATIBILITY_MODE / MIXED / UNCLEAR), with a short rationale and a confidence rating. Use as the first triage step before routing.

ADR registry discipline review for exeris-docs. Use on every ADR PR — verifies number-first reservation, filename pattern, location-by-scope, visibility taxonomy, license taxonomy, and link stubs.

Document shape classifier for exeris-docs. Use whenever a user asks to "draft an ADR / RFC / research note" — confirms Research / RFC / ADR / refactor-note is the right shape, and selects the right template.

Drift-pattern sweep review for exeris-docs. Use after any non-trivial HLA / whitepaper / large-doc edit to catch the 10 recurring drift patterns.

Router/Planner skill for exeris-docs. Produces primary agent, secondary handoffs, execution order, validation gates, and minimal next action for a docs task.

Three-tier narrative review for exeris-docs. Use whenever Tier 1 / Tier 2 / Tier 3 / Family-product framing appears in an edit, or when claims about engine swap / cap manifest / SKU runtime / Spring Runtime consumers surface.

Router/Planner triage skill for exeris-docs. Classifies task type (document shape / ADR registry / narrative integrity / large-doc edit / drift sweep), scope, severity, and recommends primary agent.

Path-sandbox review for filesystem-bound tool handlers in exeris-ai-bridge. Use on every PR touching `src/tools/docs/**` or any new filesystem-bound family.

Read-only discipline review for MCP tool handlers in exeris-ai-bridge. Use on every new/changed tool handler to refuse mutation surfaces.

Router/Planner skill for exeris-ai-bridge. Produces primary agent, secondary handoffs, execution order, validation gates, and minimal next action for an MCP server task.

Router/Planner triage skill for exeris-ai-bridge. Classifies task type (architecture / implementation / tool-family / transport / cross-repo / docs), scope, severity, and recommends primary agent based on primary risk against ADR-025.

Tool family discipline review for exeris-ai-bridge. Use whenever a new tool is added, an existing tool is renamed/moved, or a new family is proposed.

The Wall by construction review for exeris-ai-bridge. Use whenever a dep, transport, or kernel-integration change could even theoretically link the TS/Node process into the Java kernel classpath.