agentleFS
virtuoso-devvirtuoso-dev/skills/reactive-engine/SKILL.md

reactive-engine

Skill1 starsChanged 44 days agoMITView on GitHub

Manage application state with the @virtuoso.dev/reactive-engine-* package family. Use this skill when (1) defining reactive state with Cell, Stream, Trigger, or Resource nodes, (2) wiring React components through EngineProvider, useCellValue, useCellValues, or usePublisher, (3) fetching data with Query and Mutation from reactive-engine-query, (4) routing with Route, Layout, and Guard from reactive-engine-router, (5) persisting cells with linkCellToStorage, (6) architecting a component or library on top of the engine, (7) observing or troubleshooting propagation with describeNode, observeDiagnostics, useEngineDiagnostics, or useRemoteEngineDiagnostics, or (8) any task involving Engine, pub, sub, getValue, combine, pipe, link, changeWith, the e namespace, missing or duplicate emissions, unexpected distinct suppression, or the error "No active engine found".

1 other project uses a version of this.

Adopted by

Most recently taken up by

  • Other versions1 project uses a version of this.

    Word for word · 1

    1. petyosi/react-virtuoso★ 6.5k100% same

    Edited · 0

    None found.

    Wrote their own version · 0

    None found.

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository3 other files it gives its agents.

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 3 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use diagnostics when a value is missing, duplicated, unexpectedly suppressed, routed through the wrong engine, or throws during synchronous …

    line 106
  2. Very specific:

    Start with value capture disabled.

    line 110
  3. Very specific:

    Remove the observer after reproduction.

    line 113
  4. Distinctive:

    Read propagation diagnostics for the record model, safety rules, and evidence-to-cause table.

    line 115
  5. Distinctive:

    Follow the data-table architecture (full patterns):

    line 166

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Reactive Engine

A reactive state system built on a graph of typed nodes. Five packages compose:

PackageProvides
@virtuoso.dev/reactive-engine-coreNodes (Cell, Stream, Trigger, Resource, DerivedCell), Engine, operators, combinators, the e namespace
@virtuoso.dev/reactive-engine-reactEngineProvider, state hooks, local and remote diagnostic hooks
@virtuoso.dev/reactive-engine-queryQuery and Mutation data fetching on top of nodes
@virtuoso.dev/reactive-engine-routerRoute, Layout, Guard, Router — routes are nodes
@virtuoso.dev/reactive-engine-storagelinkCellToStorage (localStorage / sessionStorage / cookie)

It also powers @virtuoso.dev/data-table internally — the same concepts apply when remote-controlling the table, and data-table's architecture is the reference for building on the engine (building-on-the-engine).

Mental model: nodes are definitions, engines hold state

Node constructors return inert references (symbols at runtime), defined at module scope with a $ suffix. They hold no state. An Engine instance activates nodes lazily on first use and owns their values — two engines using the same node have independent state. This is what makes engine-based libraries reusable: one module-scope graph, one engine per component instance.

import { Cell, Stream, Engine, e } from '@virtuoso.dev/reactive-engine-core'

const count$ = Cell(0)
const engine = new Engine()
engine.sub(count$, (value) => console.log('count:', value))
engine.pub(count$, 1) // logs 'count: 1'

Node types — pick by statefulness:

NodeStateUse for
Cell(initial, distinct?)Stateful, has a current valueApp state, settings
DerivedCell(initial, source$)Stateful, tracks a sourceRead-only computed state
Stream<T>(distinct?)Stateless, emits valuesEvents, commands
Trigger()Stateless, valueless"Something happened" signals (refetch, reset)
Resource(factory)Factory-initialized per engineObjects needing setup/disposal

Nodes are distinct by default using reference equality (===): publishing the same reference (or an equal primitive) does not re-emit. Always publish new references from updates (new Map(old), spread). Pass false to always emit, or a comparator (prev, next) => boolean (true = equal = suppress); prev is undefined on first emission.

The three API flavors (most important distinction)

The same verbs exist in three forms — using the wrong one is the main source of errors:

FlavorExamplesWhereEffect
Module-scope combinatorslink, pipe, sub, changeWith, combine / e.*Anywhere, incl. module top levelDeferred wiring, applied once per engine when the nodes activate
Engine instance methodsengine.pub, engine.sub, engine.getValueWherever you hold an engineImmediate, on that engine
Context-bound utilitiesstandalone pub, getValue, pubIn / e.pub, ...Only inside subscription callbacks and node initsActs on the currently executing engine

Calling a context-bound utility elsewhere throws No active engine found. You can use getValue only in the context of node subscription callbacks. — fix by using engine.* methods, React hooks, or moving the code into a subscription. Note the asymmetry: standalone sub/link/changeWith are safe at module scope (they defer); standalone pub/getValue are not.

The e export bundles all combinators, operators, and context utilities into one namespace — idiomatic for wiring blocks:

e.link(
  e.pipe(
    toggleTheme$,
    e.withLatestFrom(theme$),
    e.map(([, t]) => (t === 'light' ? 'dark' : 'light'))
  ),
  theme$
)

Wiring the graph

  • link(source$, sink$) — forward emissions; pipe(source$, ...ops) — derive a new node
  • combine(a$, b$, ...) — emit [a, b, ...] when ANY source emits; withLatestFrom(...) (operator) — read other nodes passively, emit only when the source emits
  • changeWith(cell$, source$, (cell, value) => next) — reducer-style cell update from a stream (return a new reference)
  • sub, subMultiple, singletonSub (replaces the previous singleton subscription — use for callbacks driven by React renders)
  • Operators: map(fn, distinct?), filter (supports type guards), mapTo, scan, once, throttleTime, debounceTime, delayWithMicrotask, onNext, handlePromise

Publishing is transactional: engine.pubIn({ [a$]: 1, [b$]: 2 }) batches into one cycle; the graph executes in topological order, so diamond dependencies emit downstream exactly once with consistent inputs. Emissions are synchronous; there are no error/completion channels (exceptions abort the cycle — handle errors at the edges). Subscribing never replays a current value; read cells with getValue when needed. See transactions.

React integration

import { EngineProvider, useCellValue, usePublisher } from '@virtuoso.dev/reactive-engine-react'

const Counter = () => {
  const count = useCellValue(count$)
  const setCount = usePublisher(count$)
  return <button onClick={() => setCount(count + 1)}>{count}</button>
}
  • useCellValue(cell$) subscribes and re-renders; useCellValues(a$, b$, ...) returns a tuple with a single combined subscription (prefer it for cells that change together); useCell is the useState shape; usePublisher(node$) returns a stable publish function
  • EngineProvider props: initFn(engine) (one-time setup: register nodes, seed with pubIn, attach bridges), initWith (initial cell values), updateFn/updateDeps (re-publish changed props; use singletonSub for callback props so re-renders replace instead of stack), engineId, engineRef, and diagnostics (attach before initFn)
  • Remote hooks (useRemoteCellValue, useRemotePublisher, useRemoteCell, useRemoteCellValues, useRemoteEngineDiagnostics) reach another provider's engine via useEngineRef() or a string engineId — value hooks return undefined until that engine mounts; diagnostic observation waits and attaches automatically
  • Components stay projection-only: read cells, publish actions. Logic lives in module-scope wiring, testable with a bare Engine and no React

Troubleshoot propagation with diagnostics

Use diagnostics when a value is missing, duplicated, unexpectedly suppressed, routed through the wrong engine, or throws during synchronous propagation:

  1. Label the root, expected sink, and intermediate nodes with describeNode. Add narrow summarize functions only for application-specific values that the built-in bounded summary cannot explain.
  2. Attach at the correct lifecycle: EngineProvider.diagnostics includes initFn; useEngineDiagnostics observes the nearest engine after commit; useRemoteEngineDiagnostics is for sibling tools and follows engine availability. Treat provider initialization records as at-least-once in development because React Strict Mode can evaluate initialization more than once.
  3. Start with value capture disabled. Read roots and events in order. Inspect distinct-suppressed, no-candidate, prune events, and error.phase before enabling captureValues: 'summary'.
  4. Compare engineInstanceId and engineLabel when state appears to come from the wrong instance. Use transactionId, origin, and parentCycle for synchronous nested or parent-child propagation. Async work starts a new transaction and needs application-level correlation.
  5. Enable summaries only for the nodes needed to verify the hypothesis. Redact before exporting. Node summarizers receive live values and must not mutate them or cause side effects.
  6. Remove the observer after reproduction. Core and React retain no history; build any diagnostic UI on a bounded external store, not EngineProvider state.

Read propagation diagnostics for the record model, safety rules, and evidence-to-cause table. Read React integration for provider, local-hook, remote-hook, Strict Mode, and SSR timing.

Data fetching (reactive-engine-query)

export const tasksQuery = Query<{ listId: string }, Task[]>({
  queryFn: async ({ listId }, signal) => {
    const res = await fetch(`/api/tasks?listId=${listId}`, { signal })
    if (!res.ok) throw new Error('Failed to fetch tasks')
    return res.json()
  },
  initialParams: { listId: '' },
})

Query returns nodes: data$ (cell with { type: 'pending' | 'success' | 'error', data, error, isLoading, isFetching, ... }), params$ (publish to refetch with new params), refetch$, invalidate$ (keep showing stale data, isFetching: true, re-execute), unload$ (abort and clear), enabled$. Queries auto-execute on activation and param changes; each execution aborts the previous via the signal. Options: retry (default 3, exponential backoff), refetchInterval, initialData, enabled.

Mutation({ mutationFn, onSuccess?, onError? }) returns data$, mutate$, reset$. Mutations never auto-execute — publish params to mutate$. Wire mutations to refetch queries in the graph, not in components:

e.link(
  e.pipe(
    e.merge(createTask.data$, deleteTask.data$),
    e.filter((r) => r.type === 'success'),
    e.map(() => undefined, false) // distinct=false so consecutive successes each fire
  ),
  tasksQuery.refetch$
)

Routing (reactive-engine-router)

export const user$ = Route('/users/{id:number}', UserPage) // params typed from the path
const root = Layout('/', ({ children }) => <main>{children}</main>)

<Router routes={[user$]} layouts={[root]} guards={[authGuard]} />

Route nodes emit typed params when matched, null otherwise. Navigate by publishing params to the route node (usePublisher(user$)({ id: 42 })). Pipe route params into query params to fetch on navigation. Guard(pattern, fn) runs on matching navigations — its context offers continue(), navigate(), redirect(); async guards render the route inside Suspense. Layouts match by URL prefix and nest by specificity; LayoutSlotPortal/LayoutSlot/LayoutSlotFill let pages fill layout regions. Router syncs with browser history by default.

Persistence (reactive-engine-storage)

linkCellToStorage(theme$, { storageType: 'localStorage', key: 'app-theme', debounceMs: 300 })

storageType: 'localStorage' | 'sessionStorage' | 'cookie' (cookie adds cookieOptions). Reads happen lazily, once per engine, when the engine first activates the cell; writes are debounced; localStorage links sync across tabs. The engine's id namespaces keys (my-app:app-theme) so multiple instances persist independently. SSR-safe (no-op without window).

Building a component or library on the engine

Follow the data-table architecture (full patterns):

  • Module-scope node graph + one engine per component instance (via EngineProvider)
  • Public API = action streams in (setX$, resetX$), state cells out (xState$, often DerivedCell); wiring stays internal
  • Props flow into cells via pubIn in initFn/updateFn; callback props via singletonSub
  • Optional features as separate modules (subpath exports) that attach wiring to core nodes — unused features cost nothing because inits are lazy
  • External control via engineRef/engineId + remote hooks against the exported nodes
  • Bridge imperative subsystems with subscribe-both-ways functions whose cleanup reaches engine.onDispose

Common mistakes

  • Calling standalone pub/getValue/pubIn outside a subscription callback — use engine.* methods or React hooks.
  • Mutating and republishing the same object — distinct is reference equality; nothing emits. Publish new references.
  • Expecting a node to hold state by itself — values live in an engine; the same Cell in two engines has two values.
  • Expecting Stream subscriptions to fire on subscribe with a current value — streams are stateless; only Cell has a value.
  • Piping into a trigger through map without distinct: false — consecutive identical mapped values get filtered.
  • Triggering a Mutation by changing params — mutations only run when params are published to mutate$.
  • Plain sub in code that re-runs with React renders — subscriptions accumulate; use singletonSub.
  • Creating cycles in the graph — cycles hang or loop; break one direction with an action stream or delayWithMicrotask.

References

Guides in references/, per package:

  • references/core/ — core-concepts (node types, distinct, the e namespace), engine-and-lifecycle (activation, API flavors, child engines, disposal), transactions (cycles, topological execution, RxJS differences), operators-and-combinators (full reference), building-on-the-engine (architecture patterns from data-table), propagation-diagnostics (records, safety, troubleshooting), README
  • references/react/ — react-integration (EngineProvider lifecycle, hooks, remote hooks, SSR), README
  • references/query/ — queries-and-mutations (result states, lifecycle, wiring patterns), README
  • references/router/ — routing (path syntax, navigation, layouts, guards), README
  • references/storage/ — storage-links (timing, namespacing, cross-tab sync), README

The JSDoc on the exported symbols in @virtuoso.dev/reactive-engine-core is extensive — when in doubt about a signature, read the type definitions shipped with the package.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.