Rust async web

Rust Async And Web

Use this skill for async Rust services and full-stack Rust applications. Keep runtime, HTTP, UI, and persistence boundaries explicit so domain logic remains testable without the framework. Use hexagonal-architecture when handlers, jobs, server functions, message consumers, persistence, or external clients need formal ports/adapters or shared application use cases.

Load rust-engineering when async/web work also changes core Rust implementation such as types, ownership, traits, errors, features, modules, or macros. Do not load it for review-only async/web guidance.

Do not use this skill for API-contract-only work such as endpoint/resource shape, request/response/error envelopes, versioning, pagination, idempotency, or OpenAPI/AsyncAPI/protobuf artifacts; use api-design for those contracts. Do not use it for observability-only work such as log/metric/trace schemas, context-propagation standards, labels/cardinality, dashboards, alerts, SLOs, or telemetry sampling; use observability-engineering. Add this skill only when the task also changes Tokio, Axum, Leptos, SSR, hydration, WASM, task, or runtime behavior.

Native desktop GUI implementation is not this skill's primary scope. Use rust-desktop-gui for iced, egui/eframe, Slint, Tauri shell, and native event-loop work. Compose with this skill when the GUI change needs Tokio task, cancellation, channel, or backpressure mechanics, but never block the GUI event-loop thread.

Workflow

1. Inspect the runtime and stack: Cargo.toml features, tokio runtime setup, bacon.toml, Justfile/scripts, Axum routers, Leptos/cargo-leptos app and hydration configuration, server functions, WASM build target, stylesheet/static asset pipeline, middleware, tracing, and existing test recipes.
2. Define externally observable behavior with BDD-style acceptance criteria. Keep domain invariants in framework-independent types where practical.
3. Design task ownership, cancellation, timeouts, backpressure, shared state, and error propagation before adding handlers or components.
4. Implement narrow vertical slices: domain logic, adapter/handler/server function, UI or response behavior, and tests.
5. Verify the server side, client/WASM side, and hydration/SSR behavior with the repository's commands.

Security Review Prompts

Load security-review when async/web work touches auth, authorization, sessions or cookies, CORS/CSRF/CSP, redirects, SSR/hydration trust boundaries, server functions, uploads/downloads, path handling, request/response redaction, secrets, external-service calls, telemetry that may leak sensitive data, or artifact handling. Use threat-modeling before or during new auth middleware, request/SSR/server-function boundaries, background workers, queues, webhooks, external-service integrations, or sensitive data flows. Use dependency-supply-chain-review when Tokio/Axum/Leptos or toolchain dependency changes, generated assets, CI bootstrap, installers, or binaries raise provenance or advisory questions. Pair security-sensitive reviews with security-review-evidence when evidence includes sanitized HTTP traces, logs/spans, browser or server-function payloads, screenshots, generated assets, or test artifacts.

When To Choose Async Rust

• Use async deliberately for I/O-bound or high-concurrency work: network servers and clients, database calls, queues, workers, orchestration layers, file or process I/O with async APIs, and fan-out/fan-in workflows.
• Do not make code async by default. Synchronous Rust is usually clearer for CPU-bound algorithms, pure domain logic, simple CLIs, low-concurrency tools, and code where blocking APIs are sufficient.
• Async changes API shape. It affects trait boundaries, dependency injection, lifetimes, Send + 'static requirements for spawned work, cancellation, error propagation, and which tests need a runtime.
• Keep blocking libraries behind explicit boundaries. In an async service, wrap unavoidable blocking or CPU-heavy work with tokio::task::spawn_blocking, a dedicated thread pool, or a separate worker process instead of blocking Tokio runtime threads.
• Distinguish concurrency from parallelism. Async lets many tasks wait on I/O; it does not make CPU-heavy work faster unless that work is moved to real parallel execution.

Architecture Boundaries

• Keep domain logic pure, deterministic, and preferably synchronous. Avoid leaking Tokio types, channels, timers, task handles, database pools, request types, or framework state into the domain model.
• Put async I/O at application, infrastructure, adapter, or port boundaries. Application services/use cases may orchestrate async repositories, clients, queues, clocks, and external services while domain objects enforce rules.
• Use ports/traits for async dependencies only where the boundary earns it: repositories, external clients, queues, clocks, id generators, and background job interfaces. Do not force async into every layer because one adapter is async.
• Keep Tokio-specific details in infrastructure adapters and process wiring: runtime flavor, channels, task spawning, cancellation tokens, signal handling, tracing setup, and adapter-specific retry/timeout policy.
• For async traits, prefer the simplest option the repository supports. Native async fn in traits can work for static dispatch when the MSRV and auto-trait bounds fit; use explicit future return types, boxed futures, or async-trait when object safety, dyn dispatch, or tighter Send control matters.
• Use BDD examples at the system, inbound adapter, or use-case boundary; TDD domain rules with sync unit tests; test async application services with fakes; and test real adapters with #[tokio::test] integration tests.

Tokio Runtime, Coordination, And Cancellation

Use the Tokio runtime reference when the task requires runtime construction, spawning, coordination primitives, cancellation, or graceful shutdown. Keep the core constraints visible: never block runtime threads, give every task an owner and shutdown path, bound concurrency and queues, and treat cancellation as real control flow.

Local CPU Parallelism Routing

Compose with parallelism-engineering when an async/web application must redesign a local CPU-bound workload around data or task decomposition, partition sizing, bounded executor ownership, deterministic reduction, cancellation, or nested-parallelism control. That skill owns the parallel design; this skill owns the Tokio integration, such as keeping blocking CPU work off runtime threads and applying the chosen design through Rust runtime APIs.

Do not route ordinary Tokio scheduling, async I/O, channels, task lifetimes, timeouts, backpressure, or Axum/Leptos concurrency to parallelism-engineering; they remain here. Spark/PySpark execution belongs to data-platform-engineering, not this local parallelism handoff.

WebAssembly Routing

Compose with webassembly-engineering when async/web work includes a general WebAssembly decision: a WAT or .wasm artifact, WASI or WIT, the Component Model, host/guest contract, runtime or target selection, capability grants, or Wasm packaging and deployment. That skill owns those Wasm boundary and compatibility decisions. This skill retains Leptos, wasm-bindgen, Cargo, Rust target compilation, SSR/hydration, browser framework, host-integration, and test mechanics; the Wasm skill does not establish their source-level or framework support.

Axum

• Keep routers, handlers, extractors, state, and middleware thin. Move domain decisions into services or domain modules that can be tested without HTTP.
• Use typed extractors for inputs and State/substates for application state. Store shared state in Arc when it must be cloned into handlers.
• Convert domain and adapter errors into HTTP responses at the edge. Avoid leaking database, framework, or internal error details to clients. Use api-design when status codes, error envelopes, route shapes, or request/response payloads are contract decisions.
• Validate payload size, content type, authentication, authorization, and input shape before invoking domain behavior.
• Prefer Tower middleware or extractors for cross-cutting HTTP concerns such as tracing, auth, timeouts, compression, and request ids.
• Expose application assembly separately from listener startup so tests can construct the Router with controlled state.
• Test HTTP semantics by calling the router as a Tower service with tower::ServiceExt::oneshot; assert status, relevant headers, and a bounded or collected response body. Cover extractor rejection, middleware, error mapping, and state wiring where those behaviors matter.
• Use pure or application-service tests when routing adds no behavior. Bind a loopback listener on port 0 only when the test requires real transport, connection metadata, protocol behavior, or client/server integration.

Leptos And Axum-Leptos

For a technical-debt audit of an Axum + Leptos SSR application, read the Axum + Leptos technical-debt audit reference before selecting commands or diagnosing hydration, feature-gating, route, or multi-instance concerns.

• Decide whether behavior belongs in CSR, SSR, server functions, actions, resources, or ordinary backend routes. Keep security-sensitive work on the server.
• Server functions are trust boundaries. Validate input, authorization, and idempotency server-side even when the client UI already checks them.
• Components should keep state derivation clear. Avoid burying domain rules in signal plumbing when a normal Rust function or type can own the rule.
• SSR and hydration must render compatible markup. Avoid browser-only APIs, nondeterministic values, or time-dependent output during server rendering unless guarded for the target.
• WASM code should avoid unavailable server APIs and large unnecessary dependencies. Check crate features and target-specific modules.
• Axum-Leptos integrations should keep router state, server function context, static assets, fallback handling, and error pages explicit.
• Test framework-independent component and application logic with ordinary host Rust tests. Test SSR output, server functions, and backend routes through the server application with the repository's server-side features enabled.
• Test browser-targeted components with wasm-bindgen-test or the repository's established WASM harness when DOM-level behavior is the narrowest useful boundary. Match the runner and browser mode to the pinned toolchain.
• Test hydration, navigation, forms, and other user-visible full-stack behavior in a real browser. Load playwright-e2e when adding or changing checked-in Playwright tests.

Leptos-Use

• Consult the current Leptos-Use documentation and the repository's pinned Leptos and leptos-use versions before using an API. Consider the crate before hand-writing reusable reactive wrappers for browser events, media queries and preferences, storage, observers, timers, sensors, streams, or other Web APIs. Prefer native Leptos or direct web-sys code when the behavior is small, unsupported, or clearer without another abstraction.
• Inspect each function's documented SSR behavior; server fallbacks vary by utility. Use SSR-safe targets such as use_window() and use_document() instead of accessing browser globals during server rendering, and verify that fallback values cannot produce incorrect initial markup or hydration mismatches.
• In an SSR application, enable leptos-use/ssr through the application's server-only feature, not globally on the dependency. Inspect function-level crate features and avoid shipping unused defaults when WASM size or compile time matters, while preserving the repository's established feature policy.
• Prefer utilities that bind cleanup to the Leptos owner for listeners, observers, timers, and streams. Retain and call returned cleanup, pause, stop, or close controls when behavior must end before owner cleanup; confirm any same-thread restrictions in the selected API's documentation.
• Treat local/session storage as user-visible, origin-scoped client state, not trusted or secret storage. Account for decoding failures, unavailable storage, cross-tab updates, and hydration timing. For permissions and sensors, handle unsupported, denied, unavailable, and paused states rather than assuming a successful browser API call.

Load css-scss-styling when Leptos/Axum work touches .css, .scss, .sass, stylesheet entrypoints, Trunk/cargo-leptos style assets, class/style bindings, design tokens, responsive layout, CSS modules, or browser-visible cascade behavior. Keep this skill focused on Rust runtime, SSR, hydration, server functions, routing, and WASM constraints.

Development And Commands

Prefer repository scripts for full-stack apps because they often manage CSS, WASM, assets, environment, and services.

For an ordinary long-running Axum service, use a checked-in Bacon job when the repository adopts Bacon:

[jobs.server]
command = ["cargo", "run"]
need_stdout = true
background = false
on_change_strategy = "kill_then_restart"

Run it with bacon server. Adapt the command, package, features, environment, and watched paths to the repository. Add a custom kill command only when the application needs graceful shutdown and the command is verified for the target platform. Keep generic check/test jobs in rust-testing-quality.

When a project uses cargo-leptos, prefer its documented cargo leptos watch workflow because it coordinates server and browser/WASM builds. Do not wrap one file-watching development server inside another; use Bacon separately for focused Cargo quality jobs.

Useful direct checks include:

cargo check -p <server-package> --all-targets
cargo check -p <client-package> --target wasm32-unknown-unknown
cargo test -p <package>
cargo clippy -p <package> --all-targets -- -D warnings

For cargo-leptos projects, use cargo leptos test and configured cargo leptos end-to-end lanes when the pinned version and repository scripts support them. Verify host/SSR features separately from the hydrate wasm32-unknown-unknown build instead of assuming one command covers both.

For final confidence, run the repository's broader Rust, browser, and service tests when the change affects hydrated UI, routing, server functions, or HTTP behavior.

Common Pitfalls

• Introducing async where synchronous code is simpler and sufficient.
• Coupling domain models, value objects, or invariants to Tokio types.
• Confusing async concurrency with CPU parallelism.
• Creating nested runtimes instead of keeping the runtime at process edges.
• Detached tasks that hide panics, ignore shutdown, or continue using stale state after a request is gone.
• Ignoring JoinHandle results or losing task errors.
• Missing cancellation paths for background workers, request fan-out, or long-running orchestration.
• Blocking the async runtime with sync I/O, CPU-heavy work, thread sleeps, or long critical sections.
• Holding locks, transactions, or borrowed request data across unrelated awaits.
• Unbounded channels or queues that turn overload into memory growth.
• Mixing blocking and async I/O without an explicit adapter or spawn_blocking boundary.
• Surprising Send + 'static requirements from tokio::spawn after borrowing request-local data.
• Hiding async trait allocation, object-safety, or Send tradeoffs behind a port abstraction without documenting the reason.
• Mapping every error to 500 or every auth failure to the same response without preserving observability.
• Calling SQL or external services directly from UI components instead of a server-side boundary.
• Hydration mismatches caused by random ids, current time, locale, feature differences, or browser-only APIs during SSR.
• Sharing mutable state because it is convenient rather than because the domain requires shared mutation.

Review Checklist

• Async work has bounded lifetime, cancellation, timeout, and backpressure.
• Handler/component/server-function boundaries are thin and testable.
• Domain invariants are not duplicated across UI, HTTP, and persistence layers.
• Errors are actionable internally and safe externally.
• Shared state and locks cannot deadlock, block the runtime, or leak across unrelated requests.
• Server-rendered and client-hydrated output agree for the changed behavior.
• Tests cover the behavior at the lowest useful layer plus at least one framework boundary when routing, extraction, hydration, or server functions are part of the change.