Skip to main content

📝 n8n Workflow Update Guide

This guide details the manual updates required to align n8n workflows with the N-Stage + HITL (Human-in-the-Loop) architecture.

1. Load Batch Workflow​

Pattern: Use Postgres View​

Instead of querying the ai_tasks table directly, use the v_runnable_tasks view to ensure dependency logic is respected.

SELECT t.*, 
       tb.name as batch_name,
       tb.grade_code
FROM v_runnable_tasks t
LEFT JOIN task_batches tb ON t.batch_id = tb.id
LIMIT 20;

Required Fields​

Ensure the transformation node includes these metadata fields:

  • requires_approval: Boolean for HITL.
  • next_task_config: JSON for auto-chaining child tasks.

2. Sync Back to Supabase​

Determine Final Status​

After AI processing, check if the task requires manual approval before proceeding.

const task = $input.item.json;
const aiResult = $('Gemini API Call').item.json;

let finalStatus = 'completed';
if (task.requires_approval) {
  finalStatus = 'awaiting_approval';
}

return { 
  ...task,
  output_data: aiResult.data,
  final_status: finalStatus
};

3. Create Next Task (Auto-chaining)​

For non-HITL tasks, trigger the next step in the chain immediately via Supabase RPC.

if ($json.final_status === 'completed' && $json.next_task_config) {
  await $http.request({
    method: 'POST',
    url: `${ENV.SUPABASE_URL}/rest/v1/rpc/create_next_task_in_chain`,
    body: { p_parent_task_id: $json.supabase_task_id }
  });
}

🛠️ Implementation Checklist​

  1. Update Load Batch to use v_runnable_tasks.
  2. Update Sync Back to handle awaiting_approval.
  3. Add auto-chaining node for continuous pipelines.
  4. Test with a task where requires_approval=true.

Last Updated: 2026-02-25