n8n Automations

Overview

Travel Tamers uses n8n (v2.6.3), a self-hosted workflow automation platform, to connect its services and automate repetitive business processes. The n8n instance runs as a Docker container on the Hostinger VPS, accessible at automations.traveltamers.com behind Traefik with automatic SSL.

There are 60 designed automations (numbered A-1 through A-60) covering the full business lifecycle: from generating daily social media content with AI, to nurturing leads through email sequences, to sending pre-trip destination guides and post-trip feedback requests. Of these, 35 are currently deployed in the n8n instance. The original 60 automation IDs are the canonical reference, but some workflows were consolidated or deduplicated during deployment — the actual workflow count on the VPS is 35. The remaining automations are designed and documented, ready to be built as the business scales.

Numbering convention: All automations are referenced as A-1 through A-60 throughout the codebase, architecture docs, and n8n workflow names. The master specification lives in architecture/automation-workflows.md.

Environment variable migration (R9): 29 of the 35 deployed workflows were migrated from hardcoded values (API keys, Slack channel IDs, webhook secrets) to n8n environment variables using the $env.VARIABLE_NAME syntax. This means secrets are no longer baked into workflow JSON files — they're injected via Docker Compose at container startup.

Connected Services

The n8n instance connects to nine external services plus two internal PostgreSQL databases. Each integration is configured via environment variables in Docker Compose or directly as n8n credentials.

Architecture

The n8n container sits at the center of the Travel Tamers automation ecosystem, connecting inbound channels (LinkedIn, website forms, email replies) to outbound actions (Slack messages, social posts, CRM updates, email sends). It communicates with the TT-API via authenticated HTTP requests and connects directly to both PostgreSQL databases for queries and writes.

Webhook Endpoints

Cron Schedules

Webhook authentication: All workflows that call back to TT-API or Nexus send an X-Webhook-Secret header using $env.WEBHOOK_SECRET. Nexus validates this at /api/n8n/webhook/:action with a timing-safe comparison. See the Webhook Secrets section for the full secret topology.

Master trigger pattern: A-33 (departure date cron) is the master trigger for 12 downstream workflows. It scans all bookings daily and fires the appropriate workflow based on how many days until departure or after return: A-28 at -180d, A-29 at -90d, A-30 at -14d, A-31 at -3d, A-32 at -7d, A-35 on departure day, A-36 at +1d, A-39 at return +1d, A-40 at return +3d, A-41 at return +7d, A-45 at return +365d.

Buffer GraphQL API

All social media scheduling goes through the Buffer GraphQL API at api.buffer.com. The old v1 REST API at api.bufferapp.com is dead — it didn't support Instagram and has been fully retired. Every workflow that publishes to social media was migrated to the GraphQL endpoint during R8-R10.

API Details

Cloudflare protection: Every Buffer HTTP node must include a User-Agent: TravelTamers-n8n/1.0 header. Without it, Cloudflare blocks the request with a 403. This is a hard requirement — n8n's default user agent gets rejected.

Channel Profile IDs

Three Buffer profile IDs are set as n8n environment variables. These identify which Buffer channels receive published posts.

GraphQL Mutation Example

mutation CreatePost($input: CreatePostInput!) {
  createPost(input: $input) {
    post {
      id
      text
      scheduledAt
      channels { id name }
    }
  }
}

# Variables:
{
  "input": {
    "text": "Post content here...",
    "channelIds": ["$env.BUFFER_PROFILE_LINKEDIN"],
    "scheduledAt": "2026-03-16T14:00:00Z",
    "media": [{ "url": "https://..." }]
  }
}

Workflow Categories

All 60 automations are organized into six functional categories across three priority tiers. Tier 1 are launch blockers (day 1). Tier 2 activate within 30 days. Tier 3 build when revenue starts. Click each category to see individual workflows. Of the 60 designed, 35 are currently deployed on the VPS (some consolidation occurred during implementation).

Social Media Pipeline Deep Dive

The flagship automation. Every morning at 7am ET (or via manual/API trigger), A-1 generates a full day of social media content: five curated travel stories, each with custom images, written for three platforms — LinkedIn, Facebook, and Instagram. It combines real-time news research, AI filtering, professional photography, and AI-generated imagery into a cohesive content package delivered to Slack for human review.

Full walkthrough available: For a comprehensive guide covering every node, prompt template, and data transformation in detail, see the dedicated Social Media Content Pipeline Guide.

Node Chain (14 nodes)

Trigger

Three entry points: daily cron at 7am EST, manual button in n8n, or API trigger via webhook URL.

Fetch → Parse → Google Trends

Pulls last 24 hours of messages from #feed-travel-news Slack channel, parses headlines, then cross-references with Google Trends to find what resonates.

Claude Filter → Claude Select Top 5

Two-stage AI curation. First pass sends headlines to Ollama (num_ctx: 16384) for classification, filtering to ~20–40 candidates. Second pass uses Claude Sonnet to select the top 5 stories for Travel Tamers' tech-executive audience.

Perplexity Research (x5 parallel)

Each selected story gets real-time web research via Perplexity to add context, statistics, and angles for credibility.

Unsplash Photos + FLUX Images (x5 each)

Each story gets two image options: professional stock photo from Unsplash and a custom AI image from FLUX via Replicate ($0.01–0.05 per image, 12s sleep between calls for rate limiting).

Claude Write Posts (x5)

Claude writes platform-specific versions: LinkedIn gets thought-leadership, Facebook gets community engagement, Instagram gets visual-first captions with hashtags. Character limits enforced (LinkedIn 3000, IG 2200, FB 63206).

Assemble → Slack Summary → Slack Full Content

Posts to #content-review in two messages: summary card with all 5 headlines, then individual posts with full content and images. Reviewers approve with a checkmark emoji to trigger A-2 (Emoji Approve → Buffer Publish).

Output per Run

Error Handling

• Ollama timeout (>5 min): Retry once, then skip classification and send raw headlines to Claude.
• Perplexity failure: Skip that story, proceed with remaining.
• FLUX rate limit: Use Unsplash image only, skip AI image generation.
• Full pipeline failure: Post error to #automation-errors for manual review.

Emoji Approve → Buffer Publish

The companion workflow to the content pipeline. When a reviewer in Slack reacts to a generated post with a checkmark emoji in #content-review, this workflow picks it up, parses the post content, queues it in Buffer via the GraphQL API for LinkedIn, Facebook, and Instagram, saves it to the Nexus social_posts table, and confirms with a rocket emoji reaction.

"Active version not found" fix: After deploying workflows via the n8n CLI, the emoji approve workflow (and others) threw "Active version not found" errors on execution. This was caused by workflows being imported but not published. The fix was running n8n publish:workflow --id=ID on all 35 deployed workflows. In n8n 2.x, update:workflow --active=true is deprecated — you must use publish:workflow instead. See the n8n 2.x Changes section.

Node Chain (10 nodes)

Slack Webhook

Triggered by reaction_added event from Slack. Listens for the heavy_check_mark emoji on messages in #content-review.

Route → Respond OK → Should Process

Immediately responds with 200 OK to Slack (required within 3 seconds), then filters out duplicate reactions, bot reactions, and non-content messages.

Check Approve vs. Rewrite

Distinguishes between approval (checkmark) and rewrite request (pencil emoji). Approvals proceed to Buffer; rewrites route back to Claude for content revision.

Fetch Message → Parse Post Content

Retrieves the full Slack message via conversations.history API, then parses structured post content from message blocks — extracting platform-specific text, images, and the story title from header blocks.

Buffer GraphQL → Save to Nexus → Rocket React

Uses the Buffer GraphQL API (api.buffer.com) with the CreatePost mutation to schedule posts for LinkedIn, Facebook, and Instagram with attached images. Requires a User-Agent: TravelTamers-n8n/1.0 header to avoid Cloudflare 403 blocks. Writes the approved post to Nexus social_posts table for tracking. Adds a rocket emoji to confirm the post is queued.

Slack Approval Emojis

The Parse Post Content Regex Fix

The "Parse Post Content" node in the Emoji Approve workflow is the most critical and most-debugged piece of code in the automation stack. It must reliably extract platform-specific post text and image URLs from Slack message blocks — a format that changed several times during development.

The Problem

Slack messages use a block-based format where each platform's post appears as a section block with a bold header (e.g., *:linkedin: LinkedIn*). The original parser used simple string matching that broke when Slack's block format changed or when emoji shortcodes were used inconsistently. Posts would fail to parse, and Buffer would receive empty content.

The Solution: fix-parse-regex.py

Rather than editing the node directly in the n8n UI (which is error-prone for complex JavaScript), a Python script was created at automations/fix-parse-regex.py. This script reads a workflow JSON from stdin, replaces the "Parse Post Content" node's jsCode with a battle-tested version, strips the JSON to only API-allowed fields, and outputs the result for a PUT request to the n8n API.

# Usage: pipe workflow JSON through the fixer, then PUT to n8n API
curl -s "https://automations.traveltamers.com/api/v1/workflows/WORKFLOW_ID" \
  -H "X-N8N-API-KEY: $N8N_API_KEY" \
  | python3 automations/fix-parse-regex.py \
  | curl -X PUT "https://automations.traveltamers.com/api/v1/workflows/WORKFLOW_ID" \
    -H "X-N8N-API-KEY: $N8N_API_KEY" \
    -H "Content-Type: application/json" \
    -d @-

What the Fixed Parser Does

The replacement Code node iterates through every block in the Slack message:

• Image blocks (type === "image"): Collects all image_url values into an array.
• Section blocks: Uses flexible matching with multiple patterns per platform:
  • LinkedIn: *:linkedin: or *LinkedIn
  • Instagram: *:camera: or *Instagram
  • Facebook: *:facebook: or *Facebook
• Header blocks: Extracts the story title from header type blocks.
• Fallback: If no platform-specific blocks found, uses the raw message.text truncated to platform limits.

Key pattern: The regex new RegExp("^[*][^*]+[*]\\n?") strips the bold platform header from each section's text, leaving just the post body. This is the line that broke most often and prompted the Python fix script approach.

Iteration History

The automations/ directory contains the debug trail: emoji-raw.json, emoji-raw2.json, emoji-raw3.json, emoji-raw4.json, and their corresponding fixed versions. Each iteration addressed a different edge case in Slack's block format. The final version (emoji-with-nexus.json) adds the Nexus CRM save step.

Credentials

Three credentials are configured directly in the n8n UI. These are stored encrypted in n8n's internal SQLite database and referenced by workflows via credential ID. Actual credential values aren't stored in the codebase.

Data volume warning: Credentials are stored in n8n's encrypted database, not in environment variables. If you reset the n8n container without preserving its data volume, all three credentials will need to be re-created manually in the n8n UI.

OAuth callbacks (n8n 2.6.3): OAuth callbacks now enforce user authentication. You must be logged into the n8n UI before initiating an OAuth flow, or the callback will fail with an auth error. This is a change from n8n 1.x where OAuth callbacks were unauthenticated.

Additional API keys are passed via environment variables in Docker Compose (see Environment Variables below). These are accessed in workflows via the $env variable in Code nodes or as HTTP header values.

Environment Variables

The n8n container receives environment variables through docker-compose.production.yml, including API keys for all external services, 3 Buffer profile IDs, and 12 Slack channel IDs for routing notifications. After the R9 migration, 29 workflows reference these via $env.VARIABLE_NAME instead of hardcoding values in workflow JSON.

API Keys & Tokens

Buffer Profile IDs

Slack Channel IDs (12 vars)

Twelve Slack channels are configured by ID rather than name, ensuring workflows don't break if channels are renamed. Both Shane and Debby are invited to all channels.

n8n-specific vars: The container also receives standard n8n configuration: N8N_BASIC_AUTH_ACTIVE, N8N_BASIC_AUTH_USER, N8N_BASIC_AUTH_PASSWORD, and WEBHOOK_URL (set to https://automations.traveltamers.com).

Webhook Secrets

Two separate webhook secrets handle authentication between n8n and the backend services. They're distinct values with distinct purposes — don't mix them up.

How it works in workflows: n8n workflows reference $env.WEBHOOK_SECRET in HTTP Request nodes that call back to TT-API. For Nexus callbacks, they use the same $env.WEBHOOK_SECRET value, but Nexus validates it against its own N8N_WEBHOOK_SECRET environment variable. Both sides must have matching values for the timing-safe comparison to pass.

Secret Flow Diagram

Email Templates

Transactional email templates are designed for use with React Email and sent via Resend. The templates directory is at automations/emails/ and uses the React Email component library to build responsive HTML emails that match the Dark Sanctuary brand.

Template registry: The full list of planned email templates is documented in section 5 of architecture/automation-workflows.md. Templates cover welcome sequences, booking confirmations, pre-trip guides, feedback surveys, review requests, referral invitations, and more.

Template Categories

Tech Stack

• React Email for building templates as React components with responsive layouts.
• Resend as the email delivery service (replaces raw SMTP).
• Templates use the Dark Sanctuary color palette: navy backgrounds, off-white text, gold accents.
• The automations/package.json manages React Email dependencies.

Development

The n8n instance is accessible at automations.traveltamers.com (basic-auth protected). Here are the key concepts for working with workflows effectively.

The n8n Editor

n8n provides a visual node-based workflow builder. Each node represents an operation — an HTTP request, a code block, a database query, or a service integration. Nodes pass data along edges. You can click any node to inspect configuration, test in isolation, or view the data from the last execution.

Critical gotcha: When editing a workflow in the browser, the editor runs the version loaded in the editor, NOT the saved/active version. If you deploy a workflow update via the API, you must refresh the browser to pick up changes. Otherwise "Execute" will run the stale version.

n8n expression engine: n8n's expression engine does NOT support optional chaining (?.). If you write {{$json.contact?.email}} it will silently evaluate to an empty string. Use $json.contact.email with a preceding filter node, or flatten the data in a Code node first.

Workflow Versioning

Workflow JSONs are stored in the automations/ directory of the monorepo. When you make changes in the n8n UI, export the updated workflow JSON and commit it to the repo. This ensures the codebase remains the source of truth and workflows can be redeployed.

JSON Export/Import

# Export a workflow from n8n API
curl -s "https://automations.traveltamers.com/api/v1/workflows/WORKFLOW_ID" \
  -H "X-N8N-API-KEY: $N8N_API_KEY" > automations/my-workflow.json

# Import a workflow via CLI (inside the n8n container)
n8n import:workflow --input=/path/to/workflow.json

# Or via API
curl -X POST "https://automations.traveltamers.com/api/v1/workflows" \
  -H "X-N8N-API-KEY: $N8N_API_KEY" \
  -H "Content-Type: application/json" \
  -d @automations/my-workflow.json

# IMPORTANT: After importing, you must PUBLISH the workflow (n8n 2.x)
n8n publish:workflow --id=WORKFLOW_ID

The fix-parse-regex.py Approach

For complex Code nodes where precision matters (like the Parse Post Content node), editing in the n8n UI is error-prone. The pattern established by fix-parse-regex.py is the recommended approach: write the JavaScript in a file, then use a script to inject it into the workflow JSON and deploy via the n8n API. This gives you proper version control, syntax highlighting, and the ability to test the code outside of n8n.

Best Practice

Test Individual Nodes First

Click any node and use "Execute this node" to test it in isolation with sample data. This avoids burning API credits on full pipeline runs. A full content pipeline run costs ~$0.15 in Claude credits — use node-level testing when iterating on prompts.

Cron Workflow Safety

Workflows imported from JSON are imported as inactive by default. This is intentional — cron-triggered workflows fire immediately upon activation, so all environment variables and credentials must be verified first. Review each cron workflow's schedule, confirm it makes sense, then publish it manually.

Key Files in automations/

Slack approval emojis: The standard approval emoji set for all Slack-driven workflows is: :heavy_check_mark: ✔️ (approve), :x: ❌ (reject), :alphabet-white-question: ❓ (needs clarification). Currently only :heavy_check_mark: triggers automation; reject and question are reserved for future use.

n8n 2.x Breaking Changes

The Travel Tamers instance runs n8n 2.6.3. Several breaking changes from the 1.x → 2.x upgrade affect how workflows are managed and deployed. If you're coming from n8n 1.x documentation or tutorials, these are the gotchas that'll bite you.

CLI Command Changes

Other 2.x Changes

• OAuth callbacks enforce user auth: You must be logged into the n8n UI before starting an OAuth flow. Unauthenticated callbacks are rejected.
• Task runners removed from main image: Python/shell task runners are no longer bundled in the n8nio/n8n Docker image. If you need Python execution nodes, you must run a separate n8n/runners container.
• Stricter settings file permissions: n8n now checks file permissions on its settings/config files. Make sure the n8n user owns the data volume.
• "Active version not found" errors: Workflows imported via CLI or API that haven't been explicitly published will throw this error when triggered. After importing, always run n8n publish:workflow --id=ID.

Bulk publish after import: When deploying multiple workflows at once, you must publish each one. On the Travel Tamers instance, all 35 deployed workflows were bulk-published after the initial import to resolve "Active version not found" errors across the board.

Deployment

The n8n instance runs as a Docker container on the VPS, managed through Docker Compose. Workflows are deployed by importing JSON files into the running instance, then publishing them via the CLI.

Docker Container

Deploy Workflow (n8n 2.x Process)

Copy workflow JSON to VPS

scp workflow.json shane@76.13.120.235:/tmp/

Import into n8n container

sudo docker cp /tmp/workflow.json n8n:/tmp/
sudo docker exec n8n n8n import:workflow --input=/tmp/workflow.json

Check for duplicates

import:workflow creates a new copy — it doesn't update existing workflows. Check the n8n UI for duplicates and note the old and new workflow IDs.

Unpublish the old workflow, publish the new one

sudo docker exec n8n n8n unpublish:workflow --id=OLD_ID
sudo docker exec n8n n8n publish:workflow --id=NEW_ID

Restart n8n

sudo docker restart n8n

Don't use update:workflow: The old n8n 1.x command update:workflow --active=true/false is deprecated in n8n 2.x. It may still work but it's unreliable — use publish:workflow and unpublish:workflow instead.

Full Deployment Checklist

Verify environment variables

Ensure all env vars (API keys, Buffer profile IDs, Slack channel IDs, webhook secrets, n8n config) are set in the Docker Compose file.

Import workflow JSONs

Import all workflow JSONs using the deploy process above. Workflows are created as inactive by default.

Configure n8n credentials

Create the 3 credentials (Slack Bot Token, TT Fresh DB, Nexus CRM DB) in the n8n UI if they don't already exist.

Publish all workflows

Run n8n publish:workflow --id=ID for each workflow. Without this step, webhook and cron triggers will throw "Active version not found" errors.

Test each workflow

Open each workflow and run manually once to verify all nodes succeed with real credentials.

Monitor first automated runs

Watch #automation-errors in Slack for the first 24 hours after deployment.

Webhook URL Configuration

The WEBHOOK_URL environment variable must be set to https://automations.traveltamers.com for webhook-triggered workflows to generate correct callback URLs. This URL is used by Slack event subscriptions and all inbound webhooks.

Current status (March 2026): 35 workflows deployed and published, 3 API credentials configured, all 12 Slack channel IDs set, 29 workflows migrated to environment variables, Buffer GraphQL API fully integrated. The system has been running stable on n8n 2.6.3 since initial deployment.