Claude Skills Guide

How to Use Claude Skills with n8n Automation Workflows

n8n is a self-hostable workflow automation tool with a visual node editor. Pairing it with Claude unlocks AI-powered steps inside any automation — from processing incoming webhook data to summarizing documents and triggering downstream actions. This guide covers how to call Claude from n8n using the Anthropic API, with patterns that replicate skill behavior via system prompts.

What You Can Build

Before getting into setup, here are realistic automation patterns that combine n8n with Claude intelligence:

Why Combine Claude with n8n?

n8n excels at orchestrating connections between services—moving data from webhooks to databases, sending notifications, and triggering external APIs. However, n8n’s native nodes are limited to predefined actions. When you need your workflow to understand unstructured text, make context-aware decisions, generate code, or summarize lengthy documents, Claude adds a reasoning engine to your automation without sacrificing n8n’s visual workflow management.

Prerequisites

Step 1: Store the API Key in n8n

In n8n, go to Credentials > New Credential > HTTP Header Auth:

Step 2: Add an HTTP Request Node for the Anthropic API

Call Claude directly from n8n using an HTTP Request node:

HTTP Request node settings:

Body (JSON):

{
  "model": "claude-opus-4-6",
  "max_tokens": 1024,
  "messages": [
    {
      "role": "user",
      "content": "{{ $json.prompt }}"
    }
  ]
}

This gives you a basic Claude call. To invoke a specific skill’s behavior, add a system prompt.

Step 3: Apply Skill Behavior via System Prompts

Claude skills are markdown files that define how Claude should approach tasks. In n8n, you replicate this by including the skill’s guidance as a system prompt. For TDD review behavior:

{
  "model": "claude-opus-4-6",
  "max_tokens": 2048,
  "system": "You are a TDD-focused code reviewer. Analyze code for test coverage gaps, untested code paths, and missing error handling. Suggest concrete unit tests for each uncovered path. Be specific about test names and assertions.",
  "messages": [
    {
      "role": "user",
      "content": "Review this code:

{{ $json.code }}"
    }
  ]
}

For PDF document extraction behavior:

{
  "model": "claude-opus-4-6",
  "max_tokens": 2048,
  "system": "You are a document analyst. Extract structured information from the provided text. Return JSON with: summary (2-3 sentences), key_points (array of strings), action_items (array), and entities (people, organizations, dates mentioned).",
  "messages": [
    {
      "role": "user",
      "content": "Extract information from this document:

{{ $json.documentText }}"
    }
  ]
}

Step 4: Build a Full Workflow — PR Review Automation

Here is a concrete n8n workflow that triggers on GitHub PRs and posts a Claude review to Slack.

Nodes in order:

  1. Webhook node — receives GitHub PR webhook
  2. HTTP Request node — fetches the PR diff from GitHub API:
    • URL: https://api.github.com/repos/{{ $json.body.repository.full_name }}/pulls/{{ $json.body.number }}/files
    • Authentication: Header Auth (GitHub token)
  3. Code node — extracts the diff text: 
    const files = $input.all().map(item => item.json);
const diff = files
  .map(f => `### ${f.filename}\n${f.patch || '(binary or large file)'}`)
  .join('\n\n');
return [{ json: { diff } }];
  4. HTTP Request node (Claude) — sends diff to Anthropic API with TDD system prompt
  5. Code node — extracts response text: 
    const content = $input.first().json.content;
const text = content[0].text;
return [{ json: { review: text } }];
  6. Slack node — posts review to #code-review channel

Step 5: Handle Long Documents with Batching

n8n workflows can time out on long Claude responses. Use the Split In Batches node to process large inputs in chunks:

// Code node: split text into 3000-char chunks
const text = $input.first().json.content;
const chunks = [];
for (let i = 0; i < text.length; i += 3000) {
  chunks.push({ json: { chunk: text.slice(i, i + 3000), index: i / 3000 } });
}
return chunks;

Then loop the HTTP Request node over each chunk and merge responses with a Merge node set to Combine All.

Step 6: Error Handling and Retries

Wrap your Claude HTTP Request node in a Try/Catch using n8n’s error workflow feature. For rate limit (429) errors:

// Code node after HTTP Request
const statusCode = $input.first().json.statusCode;
if (statusCode === 429) {
  return [{ json: { retry: true, waitSeconds: 60 } }];
}
return [{ json: { retry: false, result: $input.first().json } }];

Then connect a Wait node (60 seconds) and loop back to the Claude HTTP Request node.

Useful n8n + Claude Patterns

Trigger System Prompt Persona Output
New email (IMAP) Email classifier Category and routing
File upload (S3) Document analyst Extracted structured data
Code push (GitHub) TDD code reviewer Test coverage analysis
Form submission Content validator Validation result + suggestions

Conclusion

Connecting Claude to n8n automation uses the Anthropic API directly via HTTP Request nodes. The system prompt carries the skill behavior — you define what role Claude should play for each workflow step. Start with the PR review workflow above and layer in additional system prompt personas as you build out more automation pipelines.

Built by theluckystrike — More at zovo.one

Explore Our Developer Guides

© 2026 theluckystrike Developer Guides