Imported from jeremylongshore/tons-of-skills-marketplace (skills/.curated/intercom-reference-architecture/SKILL.md). Install upstream with npx skills add jeremylongshore/tons-of-skills-marketplace --skill intercom-reference-architecture. Copyright stays with the author (MIT).

Intercom Reference Architecture

Overview

A production-ready reference architecture for Intercom integrations built on four layers — API/webhook, service, Intercom client, and infrastructure — with type-safe SDK usage, webhook processing, contact sync, and Help Center management. Use it to scaffold a new integration or to review an existing one against a known-good structure.

The layers (top to bottom): the API / Webhook layer (Express routes, webhook endpoints) calls into the service layer (contacts, conversations, articles — business logic and orchestration), which calls the Intercom client layer (a singleton intercom-client SDK wrapper with typed errors, caching, and rate limit handling), all resting on infrastructure (Redis cache, job queue, monitoring). Keeping dependencies flowing strictly downward is what prevents the circular imports and test-isolation problems listed under Error Handling.

Prerequisites

• Node.js project with TypeScript and the intercom-client npm package installed.
• An Intercom access token — the SDK authenticates every request with a Bearer token read from the INTERCOM_ACCESS_TOKEN environment variable (see Step 1). Create one under Intercom → Developer Hub → your app → Authentication. Never commit it; load it from the environment.
• For webhook verification, your app's client secret to validate the X-Hub-Signature header on inbound webhook POSTs.
• Redis (optional) if you enable the caching layer.

Instructions

Use Read/Grep to inspect the current project layout, then build each layer in order — the client layer is the dependency root for every service.

1. Client layer (src/intercom/client.ts) — a lazy singleton getClient() that reads INTERCOM_ACCESS_TOKEN once, plus an IntercomServiceError that wraps raw SDK errors into a typed, retry-aware shape. Skeleton:

   let instance: IntercomClient | null = null;
   export function getClient(): IntercomClient {
     if (!instance) {
       const token = process.env.INTERCOM_ACCESS_TOKEN;
       if (!token) throw new Error("INTERCOM_ACCESS_TOKEN required");
       instance = new IntercomClient({ token });
     }
     return instance;
   }

2. Contacts service (src/services/contacts.service.ts) — findOrCreate (search-before-create to avoid 409s), syncFromCRM, mergeLead, and a searchAll async generator for cursor pagination.

3. Conversations service (src/services/conversations.service.ts) — replyAsAdmin, addNote, closeWithMessage, and a scoped open-queue search.

4. Articles service (src/services/articles.service.ts) — Help Center article create/list, defaulting new articles to draft.

5. Wire the data flow — Intercom pushes events to your webhook router; the service layer makes API calls back and persists to your database + cache.

The full project tree, every service method, and the layer/data-flow diagrams are in the full implementation walkthrough; the complete directory layout is in project-structure.md.

Output

Applying this skill produces a layered Intercom integration:

• A src/intercom/ client layer (singleton SDK wrapper + typed errors).
• A src/services/ layer with contacts, conversations, and articles services.
• src/webhooks/, src/sync/, src/api/, and src/cache/ directories wired to the layers above.
• Per-environment config/ files and a tests/ tree with unit + integration suites.

When used to review an existing project, the output is a gap report: which layers exist, which are missing, and where dependency direction is violated.

Error Handling

Examples

Once the client and service layers exist, wiring them together is a few lines — sync a CRM user, reply to and close a conversation, page through contacts, or publish an article:

const contacts = new ContactsService();
const contact = await contacts.syncFromCRM({
  id: "crm_8842", email: "ada@example.com", name: "Ada Lovelace",
  plan: "enterprise", company: "Analytical Engines Ltd",
});

See examples.md for the full set of runnable usage snippets (conversation reply/close, paginated search, Help Center publish).

Resources

• Full implementation walkthrough — every layer, method, and diagram
• Project structure reference — complete directory tree
• Usage examples — runnable wiring snippets
• Intercom API Reference
• Articles API
• Help Center API
• intercom-client npm

Next Steps

For multi-environment configuration and deployment, see the intercom-multi-env-setup skill, which extends the config/ layer described above into per-environment credential and rate-limit management.