# Revision: Cheat Sheet and Self-Check — AI Automation with n8n

Source: https://www.geekswithgeeks.com/en/n8n-ai-automation/wrap-revision

> Review n8n concepts, AI nodes and safety habits from the whole course.

## Cheat sheet

**Core**: trigger, nodes, connections, executions; data is a list of items (`{ json }`); expressions `{{ $json.field }}`. **Webhooks**: test URL vs production URL; activate to go live. **Data**: Edit Fields, Code node (all-items vs each-item), batches with Wait. **AI**: Basic LLM Chain for one prompt, Information Extractor for JSON, Text Classifier for routing, AI Agent with model, memory and tools. **RAG**: ingest (load, split, embed, store) and query with the same embedding model. **Safety**: credentials not hard-coded, webhook auth, tool limits, human approval, iteration cap. **Ops**: persistent volume, WEBHOOK_URL, encryption key, error workflow, JSON in Git.

## Questions interviewers ask

Be ready to explain: how n8n passes data as items, the test versus production webhook URL, when to use a Basic LLM Chain versus an AI Agent, how RAG works end to end, how you would defend an AI workflow against prompt injection, and how you would keep a self-hosted instance reliable.

**Quiz:** Your webhook works when testing but never fires in production. What is the most likely cause?

- [ ] The model is too small
- [ ] JSON is broken
- [x] The workflow is not activated or the test URL is being called
- [ ] Docker is installed

*Answer:* The workflow is not activated or the test URL is being called. Production webhooks run only for active workflows and use the /webhook/ path, not /webhook-test/.

**Quiz:** You lose N8N_ENCRYPTION_KEY after moving servers. What is the consequence?

- [x] Saved credentials cannot be decrypted
- [ ] Nothing, it is optional
- [ ] Workflows run twice as fast
- [ ] The editor gets a new theme

*Answer:* Saved credentials cannot be decrypted. Credentials are encrypted with that key, so without it they are unreadable and must be re-entered.

**Quiz:** Which step belongs behind human approval in an AI email-reply workflow?

- [ ] Reading the incoming email
- [ ] Classifying the email
- [ ] Logging the token count
- [x] Sending the reply to the customer

*Answer:* Sending the reply to the customer. Sending is outward-facing and irreversible; reading and classifying are internal and safe.