Human-in-the-Loop Workflow for Elegant Frontend Training

This document describes the workflow for human-in-the-loop (HITL) tasks in the Elegant Frontend Training company.

Overview

The COO is responsible for setting up workflows that enable:

Manual Task Identification: AI agents identify tasks that require human input
Owner Alerting: Owners are notified when their attention is needed
Approval Gates: Human approval before proceeding with sensitive operations
Feedback Loops: Human feedback incorporated into AI workflows

Implemented: Agent Content Pipeline Skill

The agent-content-pipeline skill has been installed and assigned to the COO agent. This provides a complete human-in-the-loop workflow for content creation:

Directory Structure

elegant/
├── drafts/     # Agent writes drafts here
├── reviewed/   # Human reviews, gives feedback
├── revised/    # pi automatically rewrites based on feedback
├── approved/  # Human-approved, ready to post
├── posted/    # Archive after posting
└── templates/ # Platform templates

Workflow

drafts/ → reviewed/ → revised/ → approved/ → posted/
  write     review      pi         approve    post

Commands

content list - Show pending content
content review <file> - Human reviews/approves
content edit <file> - Open in editor
content post <file> - Post to platform
content thread <file> - Add feedback note

Platforms Supported

LinkedIn, X (Twitter), Reddit, dev.to, Hashnode

Paperclip Native Features

1. Status-Based Workflow

Use Paperclip issue statuses to track human involvement:

Status	Meaning
`todo`	Ready for AI processing
`in_progress`	AI is working on it
`in_review`	Awaiting human review
`done`	Completed

2. Comment-Based Alerts

AI agents can alert owners by posting comments:

@owner Please review this task. I've prepared the following options:

1. Option A: [description]
2. Option B: [description]

Please let me know which approach you'd prefer, or if you have any questions.

3. Approval Requests

For formal sign-offs, use Paperclip approvals:

Routine Notes MCP Integration

Setup

Install Routine desktop app from https://routine.co/download
Run the MCP server:
```
npx routine-mcp-server
```
Configure your AI agent to connect to the MCP server

Usage

The Routine MCP provides tools for:

Reading and writing notes
Managing tasks
Accessing calendars
Creating reminders

Alert Workflow

AI identifies a task requiring human input
AI creates a note in Routine with task details
AI creates a reminder for the owner
AI posts a comment in Paperclip alerting the owner

MCP Integration (Optional)

Routine Notes MCP

For connecting to Routine (calendars, tasks, notes):

Install Routine desktop app from https://routine.co/download
Run the MCP server:
```
npx routine-mcp-server
```
Configure your AI agent to connect to the MCP server

Notes MCP Server (Alternative)

Alternative MCP server for note management:

npx @sayranovv/notes-mcp-server

skills.sh

The skills.sh file in this directory contains commands to install additional skills:

./skills.sh

This will install:

Human-in-the-loop agent skills
Routine MCP server
Notes MCP server (alternative)

Best Practices

Clear Communication: Always explain what you need from the human
Provide Options: When possible, offer choices rather than open-ended questions
Set Expectations: Indicate when you need a response by
Document Decisions: After human feedback, document what was decided
Follow Up: If no response, follow up appropriately

Example Workflow

1. AI is working on task ELE-20 (content creation)
2. AI identifies that content needs legal review
3. AI:
   - Changes status to "in_review"
   - Posts comment: "@legal Please review the attached content for compliance"
   - Creates Routine note with content details
   - Sets reminder for legal team
4. Human reviews and responds
5. AI incorporates feedback and continues
6. AI changes status to "in_progress" or "done"