n8n Autopilot — AI Workflow Builder & Deployer

By Dr. FIRAS — LinkedIn

Build complete n8n workflows from natural language, validate them, and deploy them to your n8n instance — all in one step. No placeholders, no TODOs, no manual assembly required.

Golden Rules

ALWAYS:

1. Generate every node with real, configured parameters (URLs, paths, expressions)
2. Wire all connections explicitly in the connections object
3. Include exactly one trigger node per workflow
4. Assign node positions for a clean canvas layout (~200 px spacing)
5. Validate the workflow before presenting it to the user
6. Offer to deploy automatically when N8N_API_KEY is configured

NEVER:

• Output placeholder nodes ("configure this later", "add your logic here")
• Leave url, path, or jsCode fields empty
• Create disconnected nodes
• Present partial flows expecting manual completion

Setup

Two environment variables are needed:

Set them via OpenClaw settings (~/.config/openclaw/settings.json):

{
  "skills": {
    "n8n-autopilot": {
      "env": {
        "N8N_API_KEY": "your-key",
        "N8N_BASE_URL": "https://your-n8n.example.com"
      }
    }
  }
}

Verify connectivity:

python3 scripts/n8n_deploy.py ping

Workflow Generation Pipeline

Follow these steps for every workflow request:

Step 1 — Understand the intent

Determine four things before writing JSON:

• Trigger: webhook, schedule, manual, or external event?
• Data source: webhook body, API, database, file?
• Processing: transforms, conditionals, enrichment?
• Output: API call, notification, database write, webhook response?

If the request is clear, skip to generation. Do not over-ask.

Step 2 — Plan the node chain

Map the flow mentally:

Trigger → [Ingest] → [Transform] → [Branch?] → [Act] → [Respond/Store]

Step 3 — Generate the JSON

Use the WorkflowForge builder (see scripts/workflow_forge.py) or write raw JSON. Output a single valid JSON object ready for n8n's "Import from JSON".

Step 4 — Self-validate

Run through the validation checklist below. Fix every issue before showing the user. You can also call:

python3 scripts/workflow_inspector.py check --file workflow.json

Step 5 — Deploy (if API is configured)

python3 scripts/n8n_deploy.py push --file workflow.json
python3 scripts/n8n_deploy.py push --file workflow.json --activate

Step 6 — Explain

Provide a brief node-by-node table and import/credential instructions.

JSON Skeleton

{
  "name": "Descriptive Workflow Name",
  "nodes": [],
  "connections": {},
  "active": false,
  "settings": { "executionOrder": "v1" }
}

Node structure

{
  "id": "unique-uuid-v4",
  "name": "Human Readable Name",
  "type": "n8n-nodes-base.httpRequest",
  "typeVersion": 4,
  "position": [500, 300],
  "parameters": { ... },
  "credentials": { "httpHeaderAuth": { "id": "1", "name": "My Auth" } }
}

Connection structure

{
  "connections": {
    "Source Node Name": {
      "main": [
        [{ "node": "Target Node Name", "type": "main", "index": 0 }]
      ]
    }
  }
}

• Keys = node names (not IDs)
• main = array of arrays (outer = output index, inner = targets)
• IF nodes: index 0 = true, index 1 = false
• Terminal nodes must NOT appear as keys

Node Catalog (Quick Ref)

Triggers

Logic & Transform

HTTP & Response

Popular Integrations

AI / LLM

For full credential mapping, see references/node-catalog.md.

Expression Cheat Sheet

Validation Checklist

Before outputting any workflow, verify:

Structure:

• nodes array exists and is non-empty
• connections covers every non-terminal node
• Every node has id, name, type, typeVersion, position, parameters

Integrity:

• Exactly one trigger node
• Node names are unique
• Node IDs are unique UUIDs
• No required parameter is empty (url, path, jsCode)

Graph:

• Every connection target matches a real node name
• IF/Switch nodes define all expected output branches
• No unreachable (disconnected) nodes
• No cycles in the execution graph

Credentials:

• Nodes that need auth include a credentials block
• Credential type names are correct for their node

Expressions:

• ={{ ... }} references point to fields that upstream nodes produce

CLI Reference

Deployment (n8n_deploy.py)

# Connectivity check
python3 scripts/n8n_deploy.py ping

# List workflows
python3 scripts/n8n_deploy.py ls [--active]

# Push (create) a workflow
python3 scripts/n8n_deploy.py push --file workflow.json [--activate]

# Inspect a deployed workflow
python3 scripts/n8n_deploy.py inspect --id \x3Cwf-id>

# Activate / deactivate
python3 scripts/n8n_deploy.py on --id \x3Cwf-id>
python3 scripts/n8n_deploy.py off --id \x3Cwf-id>

# Trigger manual execution
python3 scripts/n8n_deploy.py run --id \x3Cwf-id> [--payload '{"key":"val"}']

# View recent executions
python3 scripts/n8n_deploy.py history --id \x3Cwf-id> [--limit 10]

# Execution details
python3 scripts/n8n_deploy.py exec-detail --id \x3Cexec-id>

# Execution statistics
python3 scripts/n8n_deploy.py stats --id \x3Cwf-id> [--days 7]

# Delete workflow
python3 scripts/n8n_deploy.py rm --id \x3Cwf-id>

Validation & Diagnostics (workflow_inspector.py)

# Validate a local file
python3 scripts/workflow_inspector.py check --file workflow.json

# Validate a deployed workflow
python3 scripts/workflow_inspector.py check --id \x3Cwf-id>

# Full diagnostic report (structure + performance)
python3 scripts/workflow_inspector.py diagnose --id \x3Cwf-id> [--days 14]

# Optimization suggestions only
python3 scripts/workflow_inspector.py optimize --id \x3Cwf-id>

Programmatic Builder (workflow_forge.py)

# Build from a recipe YAML and deploy
python3 scripts/workflow_forge.py build --recipe recipe.yaml --deploy

# Export as JSON
python3 scripts/workflow_forge.py build --recipe recipe.yaml --output wf.json

Optimization Heuristics

Apply when generating workflows:

1. Fan-out over chaining: independent API calls should branch in parallel, not run sequentially.
2. Batch large datasets: use splitInBatches instead of Code-node loops.
3. Error paths: add an Error Trigger or IF-based error branch for workflows with external API calls.
4. Timeouts: set executionTimeout in settings for long-running flows.
5. Avoid duplicate fetches: fetch once, reshape with Set/Code.
6. Split at 15 nodes: suggest sub-workflows via Execute Workflow node.

Output Formatting

Every workflow response must include:

1. A one-sentence summary of the workflow's purpose
2. The complete JSON in a fenced code block
3. A node explanation table:

4. Import instructions: "Open n8n → Workflows → Import from JSON → paste"
5. Credentials the user must configure post-import
6. (If API is configured) an offer to auto-deploy

Common Flow Patterns

Webhook → Transform → Respond

Webhook → Set (extract) → Code (transform) → Respond to Webhook

Schedule → Fetch → Store

Schedule → HTTP Request (fetch) → Code (process) → Google Sheets (append)

Webhook → Branch → Multi-action

Webhook → IF →  true: Slack notify  → NoOp
              false: Email alert   → NoOp

Polling with error handling

Schedule → HTTP Request → IF (ok?) →  true: Process → Store
                                    false: Slack error alert

File Layout

n8n-autopilot/
├── SKILL.md                     # This file
├── README.md                    # Project overview
├── scripts/
│   ├── n8n_deploy.py           # API client + deployment CLI
│   ├── workflow_forge.py       # Programmatic workflow builder
│   └── workflow_inspector.py   # Validation + diagnostics
└── references/
    └── node-catalog.md         # Node types & credential mapping

Troubleshooting