n8n-workflow-patterns

Use the pattern Webhook → Validate → Transform → Action → Response. Remember that incoming payload data is nested under $json.body, not $json directly, and choose between onReceived and lastNode response modes based on whether the caller needs a custom response.

Connect a language model to the AI Agent node via the ai_languageModel port, attach any node as a tool through the ai_tool port, and add Window Buffer Memory via ai_memory for conversation context. Write specific tool descriptions so the agent knows when to call each tool.

Loop with an offset, cursor, or Link-header strategy: fetch a page, extract the next cursor or increment the page counter in a Set or Code node, and loop back until no more pages remain. Add Wait nodes between requests to respect rate limits.

Webhook payload fields live under $json.body, so {{$json.email}} returns nothing while {{$json.body.email}} works. Also verify the request Content-Type header is application/json so n8n parses the body correctly.

Always use parameterized queries with placeholders like $1 and pass values through the parameters array instead of concatenating expressions into the query string. Additionally, connect with a least-privilege database user that has only the permissions the workflow needs.

Use an execution lock: at workflow start, check a Redis key, skip execution if the lock exists, otherwise set the lock with a TTL, run the workflow, and delete the lock at the end. This prevents long-running tasks from overlapping with the next scheduled run.