Docs Agent Skill

Overview

The Docs Agent is responsible for maintaining accurate, up-to-date documentation across the Unite-Hub project:

1. README.md - Project overview, installation, usage
2. .claude/claude.md - System overview and configuration
3. .claude/agent.md - Agent definitions (canonical)
4. API Documentation - Endpoint references
5. Architecture Docs - System design documents
6. Changelog - Version history and release notes

Model Configuration

Model:

claude-haiku-4-5-20251001

Reason: Fast, cost-effective, perfect for documentation tasks Cost: ~$0.10-0.20 per documentation update (vs $1-2 with Sonnet/Opus)

How to Use This Agent

Trigger

User says: "Update README", "Document new API endpoint", "Add changelog entry", "Sync docs with code"

What the Agent Does

1. Identify Documentation Need

Questions to Ask:

• What changed in the code?
• Which documentation files need updates?
• Is this a new feature, bug fix, or refactor?
• What's the target audience (developers, users, both)?

2. Read Current Documentation

Step A: Locate Documentation Files

# Find all markdown files
find . -name "*.md" | grep -v node_modules

Key Documentation Files:

• README.md

  - Main project README

• .claude/claude.md

  - System overview

• .claude/agent.md

  - Agent definitions

• ARCHITECTURE.md

  - System architecture

• API_DOCUMENTATION.md

  - API reference

• DEPLOYMENT_GUIDE.md

  - Deployment instructions

• *.md

  files in root (various guides)

Step B: Read Relevant Files

// Use text_editor tool
text_editor.view("README.md");
text_editor.view(".claude/claude.md");

3. Update Documentation

Step A: README.md Updates

README structure for Unite-Hub:

# Project Name

Brief description (1-2 sentences)

## Features

- Feature 1
- Feature 2
- Feature 3

## Tech Stack

### Frontend
- Next.js 16
- React 19
- Tailwind CSS

### Backend
- Supabase
- Next.js API Routes

### AI
- Claude Opus 4
- Claude Sonnet 4.5

## Installation

Step-by-step installation instructions

## Usage

How to use the application

## API Documentation

Link to API docs

## Deployment

Deployment instructions

## Contributing

How to contribute

## License

License information

Update Process:

1. Read current README
2. Identify outdated sections
3. Update relevant sections
4. Preserve structure
5. Maintain consistent tone

Step B: API Documentation Updates

When new API endpoint is created:

## API Endpoints

### POST /api/contacts/bulk-update

**Description**: Update multiple contacts in bulk

**Authentication**: Required (Supabase Auth)

**Request Body**:
```json
{
  "workspaceId": "uuid",
  "contactIds": ["uuid1", "uuid2"],
  "updates": {
    "status": "prospect",
    "tags": ["warm", "interested"]
  }
}

Response:

{
  "success": true,
  "updated": 2,
  "contacts": [...]
}

Errors:

• 400

  - Invalid input

• 401

  - Unauthorized

• 500

  - Server error

Example:

const res = await fetch("/api/contacts/bulk-update", {
  method: "POST",
  headers: { "Content-Type": "application/json" },
  body: JSON.stringify({
    workspaceId: "...",
    contactIds: ["..."],
    updates: { status: "prospect" }
  })
});

const data = await res.json();

**Step C: .claude/claude.md Updates**

After significant code changes:

1. **Update "Current System Status" section**:
   - Move fixed issues from "Critical Issues" to "Working Components"
   - Add new issues if discovered
   - Update component counts (e.g., "21 dashboard pages" → "22 dashboard pages")

2. **Update "Tech Stack Details" section**:
   - Add new dependencies
   - Update version numbers
   - Document new integrations

3. **Update "Data Flow Analysis" section**:
   - Add new workflows
   - Fix broken flows
   - Document critical break points

4. **Update "Immediate Next Steps" section**:
   - Remove completed tasks
   - Add new priority tasks
   - Re-prioritize remaining work

**Step D: .claude/agent.md Updates**

When agents change:

1. **Add new agent definition**:
   ```markdown
   ### 7. New Agent Name

   **ID**: `unite-hub.new-agent`
   **Model**: `claude-sonnet-4-5-20250929`
   **Skill File**: `.claude/skills/new-agent/SKILL.md`

   #### Role
   What this agent does

   #### Capabilities
   - Capability 1
   - Capability 2

   #### Tools Available
   - Tool 1
   - Tool 2

   #### When to Invoke
   - Scenario 1
   - Scenario 2

2. Update agent interaction patterns if workflows change

3. Update tool usage guidelines if new tools are added

4. Generate Changelog Entries

When code changes are released:

# Changelog

## [1.1.0] - 2025-11-16

### Added
- New bulk update API endpoint for contacts
- Dashboard analytics page
- Hot leads email send functionality

### Changed
- Updated README with new API documentation
- Improved workspace filtering across all pages

### Fixed
- Fixed missing workspace filter in dashboard overview
- Fixed undefined organization in AuthContext
- Fixed missing import in src/lib/db.ts

### Security
- Re-enabled authentication on all API routes
- Added RLS policies for new tables

Changelog Categories:

• Added: New features
• Changed: Changes to existing features
• Deprecated: Features that will be removed
• Removed: Removed features
• Fixed: Bug fixes
• Security: Security improvements

5. Update Architecture Diagrams

When system architecture changes:

Text-Based Diagrams (using Markdown):

## System Architecture

┌─────────────────┐ │ Next.js App │ │ (React 19 + │ │ App Router) │ └────────┬────────┘ │ ┌────┴────┐ │ │ ┌───▼───┐ ┌──▼─────────┐ │ API │ │ Dashboard │ │Routes │ │ Pages │ └───┬───┘ └────────────┘ │ ┌───▼────────────────┐ │ AI Agent Layer │ │ • Email Agent │ │ • Content Agent │ │ • Orchestrator │ └───┬────────────────┘ │ ┌───▼────────────────┐ │ Supabase Layer │ │ • PostgreSQL │ │ • Row Level Sec │ │ • Real-time Subs │ └────────────────────┘

Mermaid Diagrams (for complex flows):

## Data Flow

```mermaid
graph TD
    A[User Login] --> B[OAuth]
    B --> C[Supabase Auth]
    C --> D[Dashboard]
    D --> E[Load Contacts]
    E --> F[Display Hot Leads]

## Common Tasks

### Task 1: Document New API Endpoint

**Steps**:
1. Read the API route file (`src/app/api/new-endpoint/route.ts`)
2. Extract:
   - HTTP methods (GET, POST, etc.)
   - Request body schema
   - Response format
   - Error codes
   - Authentication requirements
3. Add to `API_DOCUMENTATION.md`
4. Update README if it's a major feature
5. Add example usage

**Example Output**:
```markdown
### POST /api/agents/analyze

Analyze a contact using Claude AI to extract insights

**Request**:
```json
{
  "workspaceId": "uuid",
  "contactId": "uuid",
  "analysisType": "intelligence" | "sentiment"
}

Response:

{
  "success": true,
  "analysis": {
    "score": 85,
    "insights": "...",
    "recommendations": ["..."]
  }
}

### Task 2: Update README After Feature Launch

**Steps**:
1. Read current README
2. Add feature to "Features" section
3. Update "Tech Stack" if new dependencies
4. Add usage example if user-facing
5. Update screenshots if UI changed
6. Verify all links still work

**Example Update**:
```markdown
## Features (UPDATED)

### 🤖 AI Intelligence Layer
- **Email Agent**: Automatically processes incoming emails, extracts intents, analyzes sentiment
- **Content Generator**: Creates personalized marketing content using Claude Opus 4 with Extended Thinking
- **Contact Intelligence**: AI-powered lead scoring (0-100) based on engagement, sentiment, and behavior
- **Orchestrator**: Coordinates multi-agent workflows for complex automation
- **NEW: Bulk Contact Analyzer**: Analyze multiple contacts simultaneously with Claude AI ✨

### 📧 Email Integration (UPDATED)
- **Gmail OAuth 2.0**: Secure connection to Gmail accounts
- **Email Sync**: Automatic import of emails with sender extraction
- **Open/Click Tracking**: Pixel-based tracking for email engagement
- **Thread Management**: Organize emails by conversation threads
- **NEW: Email Templates**: Pre-built templates for common scenarios ✨

Task 3: Create Architecture Documentation

Steps:

1. Review codebase structure
2. Identify major components
3. Map data flows
4. Document integration points
5. Create diagrams
6. Write narrative explanation

Example Output:

# Unite-Hub Architecture

## Overview

Unite-Hub follows a **three-tier architecture**:
1. **Frontend**: Next.js 16 with React 19 (App Router)
2. **Backend**: Next.js API Routes (serverless)
3. **Database**: Supabase PostgreSQL with RLS

## Component Layers

### 1. Presentation Layer (Frontend)

**Location**: `src/app/`, `src/components/`

**Responsibilities**:
- Render UI using React 19 Server Components
- Handle user interactions
- Manage client-side state (React Context)
- Call backend APIs

**Key Components**:
- Dashboard pages (`src/app/dashboard/*`)
- UI components (`src/components/ui/*`)
- Client components (`src/components/client/*`)
- Auth context (`src/contexts/AuthContext.tsx`)

### 2. Application Layer (Backend)

**Location**: `src/app/api/`, `src/lib/`

**Responsibilities**:
- Handle API requests
- Authenticate users
- Query database
- Call external APIs (Claude, Gmail)
- Return JSON responses

**Key Components**:
- API routes (`src/app/api/*`)
- Database wrapper (`src/lib/db.ts`)
- Supabase client (`src/lib/supabase.ts`)
- Agent logic (`src/lib/agents/*`)

### 3. Data Layer (Database)

**Location**: Supabase PostgreSQL

**Responsibilities**:
- Store application data
- Enforce data integrity
- Apply Row Level Security
- Provide real-time subscriptions

**Key Tables**:
- User management: `users`, `organizations`, `workspaces`
- CRM: `contacts`, `emails`, `interactions`
- Campaigns: `campaigns`, `drip_campaigns`, `campaign_steps`
- AI: `generatedContent`, `aiMemory`, `auditLogs`

## Data Flow Example

### Contact Creation Flow

1. User fills form in Dashboard (`src/app/dashboard/contacts/new`)
2. Frontend calls API (`POST /api/contacts`)
3. API validates input, checks auth
4. API inserts contact with workspace filter
5. Supabase checks RLS policy
6. Database stores contact
7. API returns success + contact data
8. Frontend updates UI
9. Audit log created

## Integration Points

### External Services

- **Supabase**: Auth, Database, Storage
- **Claude AI**: Content generation, email analysis
- **Gmail API**: Email sync, sending
- **Stripe**: Billing (future)
- **Vercel**: Hosting, serverless functions

Task 4: Sync Docs After Major Refactor

Steps:

1. Review all changed files
2. Identify impacted documentation
3. Update technical details
4. Fix broken examples
5. Update version numbers
6. Generate changelog entry

Example Checklist:

## Documentation Sync Checklist

After refactoring `src/lib/db.ts`:

- [x] Update README.md - Database wrapper section
- [x] Update ARCHITECTURE.md - Data access layer
- [x] Update API_DOCUMENTATION.md - Database query examples
- [x] Update .claude/claude.md - Database schema section
- [x] Add CHANGELOG.md entry - "Refactored database wrapper"
- [x] Update inline code comments in `src/lib/db.ts`
- [ ] Record video walkthrough (optional for major changes)

Documentation Best Practices

Writing Style

Tone: Professional, concise, helpful

✅ Good Examples:

• "This endpoint creates a new contact in the specified workspace"
• "Returns 401 if user is not authenticated"
• "Use the

  workspaceId

  parameter to filter results"

❌ Bad Examples:

• "This is a super cool API that does amazing things!" (too casual)
• "The endpoint performs operations" (too vague)
• "Just call the API" (not helpful)

Code Examples

Always include:

• Input parameters
• Expected output
• Error handling
• TypeScript types (if applicable)

✅ Good Example:

// Fetch contacts with workspace filter
const { data, error } = await supabase
  .from("contacts")
  .select("*")
  .eq("workspace_id", workspaceId);

if (error) {
  console.error("Error fetching contacts:", error);
  return;
}

console.log(`Found ${data.length} contacts`);

❌ Bad Example:

// Get contacts
const data = await supabase.from("contacts").select("*");

Documentation Structure

Every documentation file should have:

1. Title - Clear, descriptive
2. Overview - What this document covers
3. Table of Contents - For long documents
4. Sections - Organized by topic
5. Examples - Practical code samples
6. Troubleshooting - Common issues
7. References - Links to related docs

Keeping Docs in Sync

Automated Checks (future):

# Script to check if docs are outdated
npm run docs:check

# Finds:
# - API endpoints mentioned in docs but not in code
# - Code examples with outdated imports
# - Broken internal links

Manual Review:

• Review docs after every major PR
• Update version numbers before release
• Test all code examples
• Verify all links work

Version 1 Constraints

What We Document for V1:

• ✅ All API endpoints (104 routes)
• ✅ Agent definitions and workflows
• ✅ Database schema and RLS policies
• ✅ Installation and deployment
• ✅ Architecture overview

What We Do NOT Document for V1:

• ❌ Video tutorials
• ❌ Interactive API explorer
• ❌ Automated API docs generation
• ❌ Multi-language docs
• ❌ Contributor guidelines (detailed)

Tools and Automation

Documentation Tools

Recommended (future):

• TypeDoc - Generate docs from TypeScript
• Docusaurus - Documentation website
• Mermaid - Diagram generation
• Markdown linters - Ensure consistency

Currently Using:

• Manual Markdown files
• Text editor tool for updates
• Git for version control

Documentation Templates

API Endpoint Template:

### [METHOD] /api/path

**Description**: Brief description

**Authentication**: Required | Optional | None

**Request Body**:
```json
{
  "param": "value"
}

Response:

{
  "success": true,
  "data": {}
}

Errors:

• 400 - Bad request
• 401 - Unauthorized
• 500 - Server error

Example:

// Code example

## Key Points

- **Update docs immediately** - Don't let docs drift from code
- **Use code examples** - Show, don't just tell
- **Keep it concise** - Developers skim docs
- **Link generously** - Connect related documentation
- **Test examples** - All code examples should run
- **Version appropriately** - Track docs alongside code versions

---

## Integration with Other Agents

The Docs Agent works with:
- **Frontend Agent** - Documents UI components
- **Backend Agent** - Documents API endpoints
- **Orchestrator** - Coordinates doc updates after workflows
- **All Agents** - Receives change summaries for documentation