When to use WebSocket vs SSE vs polling, reconnection with exponential backoff and jitter, heartbeat/ping-pong to detect dead connections, message ordering and deduplication, backpressure handling for slow consumers, connection state management (connecting/connected/disconnected/reconnecting), graceful degradation to polling, room/channel management, authentication on WebSocket upgrade, and SSE vs WebSocket choice criteria. Use when building real-time features like live updates, notifications, chat, dashboards, collaboration, or order tracking. Also triggers when reviewing existing real-time code for reliability gaps like missing reconnection, dropped messages, or silent connection death.

Deliver real-time stock quotes and market data with PubNub

Review the surface of a Kotlin API you're designing or exposing — a function, a class, a module, or a published library — against the concerns that govern good API design: simplicity, readability, consistency, predictability, debuggability, testability, and (for published surfaces) backward compatibility, multiplatform, and documentation. Use when the user is designing or reviewing an API — phrases like "review this API," "is this API idiomatic," "design this interface," "review my public API," "will this break binary compatibility," "API design check," or when changing a type that other modules or external consumers depend on. This is the design-and-review counterpart to the always-on idiom rules: those govern how to write a line of Kotlin, this governs how to shape and expose an API.

Convert JUnit-style test classes (`@Test` methods, `assertEquals`, `assertThrows`) to idiomatic Kotest specs (`DescribeSpec` / `BehaviorSpec`, `shouldBe`, `shouldThrow`). Maps assertions, rewrites imports, preserves test intent, then verifies the conversion by delegating to `scripts/verify-no-junit-assertions.sh`. Use when the user wants to migrate a JUnit test file to Kotest, when a test file mixes JUnit and Kotest assertions, or when a new contributor writes JUnit-style assertions in a Kotest project.

Replace java.util.Optional usage (Optional.of, Optional.empty, Optional.ofNullable, orElse, ifPresent, etc.) with idiomatic Kotlin nullable types using the question-mark suffix and the safe-call, elvis, and let operators. Strips Java's Optional workaround out of Kotlin code where the language has a better answer. Use when the user asks to "remove Optional," "kotlinify nullables," "strip Optional wrappers," or shows code that wraps nullable values in Optional for no benefit.

Use when creating, debugging, or configuring Apache Airflow 3 DAGs — builds data pipelines with TaskFlow API or traditional operators, configures scheduling and asset-driven triggers, wires XCom data passing, sets up sensors and deferrable operators, generates dynamic task mappings, and structures multi-layer test suites. Triggers on DAG authoring, TaskFlow API, operators, sensors, scheduling, assets, dynamic tasks, XCom, or pipeline testing.

Decompose problems into pipelines of data transformations. Refactors loops into map/filter/reduce chains, converts nested/OO logic into composable function sequences, designs multi-step data transformation pipelines. Trigger on: "transformational programming", "data pipeline", "function pipeline", "pipe operator", "|>", "stream processing", "chained transformations", "Unix pipes", "dataflow", "decompose into steps", "write this as a pipeline", "compose functions", "chain of transformations", or restructuring imperative/OO code into data transforms. NOT for ETL infrastructure or stream processing frameworks (Kafka, Flink) — focuses on code-level function composition and transformation design patterns.

Open and warm up a JavaCV OpenCVFrameGrabber reliably on macOS, probe for real (non-black) frames before starting the main loop, and skip virtual cameras (Insta360 Link, Snap, OBS, Continuity Camera) that hijack low indices. Use when an OpenCVFrameGrabber call succeeds but returns black/stale frames, when switching between built-in and USB webcams, or when the first ~5 seconds of a pipeline produce zero face detections.

Run expensive per-frame inference (face recognition, emotion classification, ViT) at a fraction of the capture rate so the producer loop stays at 30 fps. Includes the 4× downscale pattern for Haar face detection, persisted-overlay technique for skipped frames, and Flow.sample() vs manual modulo approaches. Use when designing a vision pipeline that combines high-rate capture (30 fps+) with heavy per-frame work and you don't need every frame to be inferred.

Generate, validate, and render JSCAD v2 CAD scripts using Claude LLM. Follows DDD and swamp extension model design principles. Use when the user asks to generate a CAD model from a natural language prompt, fix a JSCAD script, or validate an STL output. Triggers on: "generate cad", "create 3d model", "jscad script", "make a part", "design a", "cad model", "generate stl", "fix jscad", "validate stl", "render model", "text to cad".

Enforce strict three-layer architecture: thin HTTP routes, pure service logic, isolated data access. Use when building or refactoring any web API — when route handlers contain database queries, when business logic is mixed with HTTP parsing, when services reference req/res, when error handling crosses layer boundaries, or when you can't test logic without spinning up a server.

Use when designing, reviewing, or implementing HTTP APIs — error and warning handling, resource state and lifecycle, read-endpoint structure, pagination, and authentication. Triggers on error responses and formats, response envelopes, webhook payloads, how an endpoint should fail; modelling a resource lifecycle (status fields, state machines, webhook event names, enum vs parseable string); structuring read endpoints (screen-shaped/BFF vs canonical resource, aggregation, cursor vs offset pagination); and auth design (security schemes, API keys vs bearer tokens, stepped-up tokens). Apply whenever an API surfaces a failure, state change, view of data, or auth requirement to a client.

Sanitize and validate user input at system boundaries — prevent XSS, SQL injection, path traversal, command injection, SSRF, and malformed data. Use when building any web app that accepts user input, when reviewing input handling, or when a security scan flags injection vulnerabilities. Covers frontend, backend, file handling, and shell safety.