Agent Skills: Apify Reference Architecture

Apify Reference Architecture

Overview

Production-ready architecture patterns for applications built on Apify. Three patterns scale from a single scraper to a full-stack integration:

1. Standalone Actor — one scraper deployed to the Apify platform.
2. Multi-Actor Pipeline — a discover → scrape → transform chain of Actors.
3. Full-Stack Integration — an application using Apify as a data source behind a service layer.

This skill helps you choose the right pattern, lay out the directory structure, and wire the skeleton code. Full directory trees, diagrams, and code for every pattern live in references/architecture-patterns.md; the service layer, configuration loader, and health check live in references/implementation.md.

Prerequisites

• Runtime: Node.js >=18, TypeScript, and the Apify CLI (npm i -g apify-cli).
• Packages: apify + crawlee (inside an Actor), apify-client (calling Actors from an app), zod (input validation).
• Auth: an Apify API token. Set APIFY_TOKEN in the environment; the Apify SDK and apify-client read it automatically (or pass it explicitly to new ApifyClient({ token })). Never hardcode the token — inject it via env var and validate at startup.
• Access: Read and Grep the target repository so you can match the recommended layout against the code already on disk before proposing changes.

Instructions

1. Pick the pattern. One scraper → Pattern 1. A staged workflow that discovers, scrapes, then cleans → Pattern 2. An app that consumes scraped data → Pattern 3.
2. Grep the existing repo for apify, apify-client, and Actor.main to see what is already wired, so you extend rather than duplicate structure.
3. Lay out the directory from the pattern's tree in references/architecture-patterns.md. Keep routing, extraction, and validation in separate modules.
4. Add typed input validation with zod (see src/types.ts in the reference) so bad input fails fast at the Actor boundary instead of mid-crawl.
5. Isolate every Apify call behind a service layer (Pattern 3) using the ApifyService class in references/implementation.md — the rest of the app never imports apify-client directly.
6. Load configuration once at startup via loadConfig() and layer per-environment overrides on a single base object; validate required env vars before serving traffic.
7. Expose an Apify health check so a bad token or platform outage surfaces before a user-facing scrape fails.

Output

Applying this skill produces an architecture, not a running command. Expect:

• A recommended directory layout for the chosen pattern.
• Skeleton TypeScript modules (main.ts, types.ts, service layer, config loader, health check).
• A per-environment configuration strategy and an Apify health signal.
• For pipelines, an orchestrator that reports per-stage item counts and total USD cost, e.g.:

=== Pipeline Summary ===
Discovered: 320 URLs
Scraped:    298 items
Clean:      271 items
Total cost: $0.4120

Error Handling

| Issue | Cause | Solution | |-------|-------|----------| | Circular dependencies | Service imports service | Use dependency injection | | Missing config | Env var not set | Validate at startup with loadConfig() | | Pipeline stage failure | Actor crash mid-pipeline | Add retry logic per stage | | State management | Tracking run status | Use webhook handler + database | | Run not ready error | Fetching results before SUCCEEDED | Poll getRunStatus or use a completion webhook |

Examples

Standalone Actor entry point — the minimal skeleton; full file in references/architecture-patterns.md:

// src/main.ts
import { Actor } from 'apify';
import { CheerioCrawler } from 'crawlee';
import { router } from './routes/listing';
import { validateInput, ScraperInput } from './types';

await Actor.main(async () => {
  const input = validateInput(await Actor.getInput<ScraperInput>());
  const crawler = new CheerioCrawler({
    requestHandler: router,
    maxRequestsPerCrawl: input.maxItems ?? 100,
    maxConcurrency: input.concurrency ?? 10,
  });
  await crawler.run(input.startUrls.map(s => s.url));
});

Calling an Actor from an app — via the service layer:

const apify = new ApifyService(process.env.APIFY_TOKEN!);
const { runId } = await apify.startScrape(['https://example.com']);
const results = await apify.getResults<ProductOutput>(runId);

More: the multi-stage pipeline orchestrator and the full ApifyService class are in references/architecture-patterns.md and references/implementation.md. For multi-environment setup, see the companion apify-deploy-integration skill.

Resources

• references/architecture-patterns.md — full trees, diagrams, and code for all three patterns
• references/implementation.md — service layer, config loader, health check
• Apify Platform Architecture
• API Client Reference
• Actor Development Best Practices