Usage Guide - Cybersecurity News Agent

---
noteId: "af008f20bb0711f08672f5244ab6395e"
tags: []

---

# Usage Guide - Cybersecurity News Agent

## 🎯 Quick Start

### Basic Usage (Default: Thought-Leader Style)
```bash
python agent.py
```

### With Specific Style
```bash
# Short and punchy (40-140 words)
python agent.py short

# Thought leader (150-350 words) - DEFAULT
python agent.py thought-leader

# Technical deep-dive (80-220 words)
python agent.py technical

# Personal story angle (60-180 words)
python agent.py personal-story
```

## 📝 Post Styles Explained

### 1. **Short** (40-140 words)
**Best for:** Quick updates, breaking news, hot takes

**Format:**
- Punchy hook
- 2-4 short sentences
- Direct CTA
- 2-3 hashtags

**Example use case:** "Major vulnerability just dropped - need to inform network quickly"

### 2. **Thought-Leader** (150-350 words) ⭐ DEFAULT
**Best for:** Industry insights, trend analysis, strategic thinking

**Format:**
- One-line headline (5-8 words)
- 2-4 short paragraphs
- Mini-framework or example
- Concrete takeaway
- CTA
- 3-5 hashtags
- Image idea included

**Example use case:** "Weekly cybersecurity roundup with strategic insights"

### 3. **Technical** (80-220 words)
**Best for:** Engineering audience, technical analysis, tool tips

**Format:**
- Start with metric/result
- Method/context (high-level)
- Implications
- Practical mitigation
- CTA
- 3-4 technical hashtags

**Example use case:** "Deep dive for security engineers and researchers"

### 4. **Personal-Story** (60-180 words)
**Best for:** Lessons learned, career insights, relatable experiences

**Format:**
- Single-sentence scene
- Challenge + turning point
- Short lesson
- CTA
- 3 hashtags

**Example use case:** "Sharing a security incident response experience"

## 🎨 AI Image Generation

The agent now generates **3 AI image prompts** for each post that you can use with:
- **DALL-E** (OpenAI)
- **Midjourney** (Discord)
- **Stable Diffusion** (Local/Cloud)
- **Leonardo.AI** (Web)
- **Firefly** (Adobe)

### How to Use Image Prompts

1. **Run the agent** - it will generate 3 prompts at the end
2. **Copy a prompt** from the terminal or text file
3. **Paste into your AI image tool**
4. **Generate the image**
5. **Download and attach to LinkedIn post**

### Example Image Prompt
```
PROMPT 1: Professional isometric 3D illustration of a digital fortress 
with glowing blue shields protecting servers, red warning symbols 
representing threats, cinematic lighting, corporate tech aesthetic, 
4K quality, trending on Artstation
```

## 📤 Complete Workflow

### Step-by-Step Process

1. **Start Ollama** (in one terminal)
   ```bash
   ollama serve
   ```

2. **Run Agent** (in another terminal)
   ```bash
   cd /Users/kali/Codes/AI_Agent
   python agent.py thought-leader
   ```

3. **Review Output** in terminal:
   - News summary
   - AI analysis
   - Generated LinkedIn post
   - AI image prompts

4. **Generate Image**:
   - Go to DALL-E, Midjourney, or Stable Diffusion
   - Copy one of the 3 image prompts
   - Generate image
   - Download

5. **Post to LinkedIn**:
   - Open text file: `linkedin_post_TIMESTAMP.txt`
   - Copy the post content
   - Attach generated image
   - Publish!

## 📁 Output Files

After each run, you get:

### 1. JSON Output (`output_TIMESTAMP.json`)
Complete structured data:
```json
{
  "timestamp": "20251106_143022",
  "news_count": 15,
  "news_items": [...],
  "analysis": "...",
  "linkedin_post": "...",
  "ai_image_prompts": "..."
}
```

### 2. Text File (`linkedin_post_TIMESTAMP.txt`)
Ready-to-copy format:
```
======================================================================
LINKEDIN POST
======================================================================

[Your post content here]

======================================================================
AI IMAGE PROMPTS
======================================================================

PROMPT 1: [Image prompt 1]
PROMPT 2: [Image prompt 2]
PROMPT 3: [Image prompt 3]
```

## 🛡️ Safety Features

The agent includes built-in safety rules:

✅ **Never generates:**
- Exploit code
- Step-by-step attack instructions
- Active vulnerability PoCs
- Credentials or PII
- Detailed attack tool configs

✅ **Always provides:**
- High-level security concepts
- Mitigation strategies
- Responsible disclosure focus
- Professional tone
- Safe, actionable advice

## ⏰ Automated Scheduling

### Run Daily at 9 AM
```bash
python scheduled_agent.py
```

### Customize Schedule
Edit `scheduled_agent.py`:

```python
# Daily at 2 PM with technical style
schedule.every().day.at("14:00").do(lambda: run_agent_job("technical"))

# Every Monday at 9 AM with thought-leader style
schedule.every().monday.at("09:00").do(lambda: run_agent_job("thought-leader"))

# Every 6 hours with short style
schedule.every(6).hours.do(lambda: run_agent_job("short"))
```

## 🎯 Pro Tips

### 1. **Mix Styles Throughout the Week**
- Monday: `thought-leader` (big picture insights)
- Wednesday: `technical` (deep dive)
- Friday: `short` (quick round-up)

### 2. **Review Before Posting**
- Always read the generated post
- Add personal touches
- Verify facts if specific numbers are mentioned
- Adjust tone to match your brand

### 3. **Image Selection**
- Try all 3 prompts to see which image works best
- Adjust colors to match your LinkedIn brand
- Use consistent visual style across posts

### 4. **Engagement Optimization**
- Post during peak hours (8-10 AM, 12-2 PM, 5-7 PM)
- Respond to comments promptly
- Use the CTA to drive discussion

### 5. **Content Calendar**
- Run agent daily, save outputs
- Review weekly, select best posts
- Schedule posts in advance
- Track what styles perform best

## 🔧 Troubleshooting

### Post Too Generic?
- Edit the analysis step manually
- Add specific examples from your experience
- Personalize with your insights

### Image Prompts Not Working?
- Try different AI image tools
- Simplify complex prompts
- Add "professional" or "corporate" to style
- Remove overly technical terms

### Style Not Quite Right?
- Mix and match: Generate multiple styles, combine best parts
- Edit the system prompt in `agent.py` for your preferences
- Add your personal voice in post-processing

## 📊 Example Output Comparison

### Short Style
```
🚨 Chrome just patched 3 zero-days—one actively exploited.

Update now if you haven't. These aren't theoretical risks. 
Attackers are already using them.

What's your patch management process look like?

#CyberSecurity #InfoSec #ZeroDay
```

### Thought-Leader Style
```
Headline: The Real Cost of Delayed Patches

Post:
This week's Chrome zero-days remind us: patch management isn't 
just IT housekeeping—it's strategic risk management.

[2-3 more paragraphs...]

CTA: How is your organization balancing patch speed vs. stability?

Hashtags: #CyberSecurity #RiskManagement #InfoSec #ZeroDay #CISO

Image idea: Split-screen showing unpatched system with red alerts 
vs. secure patched system with green checkmarks
```

### Technical Style
```
Chrome patches reduced exploit surface by 47% this quarter.

The three CVEs (CVE-2025-xxxx) targeted V8 engine, WebGPU, and 
file handling. High-severity ratings justified. Priority patching 
recommended within 48 hours per CISA guidelines.

[More technical details...]

#Vulnerabilities #Patches #WebSecurity #Chrome
```

### Personal-Story Style
```
I was in a security review when the Chrome zero-day alert hit.

We had 10,000 endpoints to patch. The team rallied, automated 
deployment, and covered 94% in 36 hours. Not perfect, but we 
moved fast.

Lesson: Your incident response is only as good as your 
preparation.

How do you handle emergency patches?

#CyberSecurity #IncidentResponse #InfoSec
```

## 🚀 Advanced Usage

### Custom News Time Window
```python
# Get news from last 48 hours
python -c "from agent import CyberSecurityNewsAgent; CyberSecurityNewsAgent().run(hours=48, style='technical')"
```

### Python Integration
```python
from agent import CyberSecurityNewsAgent

agent = CyberSecurityNewsAgent()
result = agent.run(hours=24, style="thought-leader")

# Access components
print(result['post'])
print(result['image_prompts'])
print(result['analysis'])
```

## 📞 Need Help?

- Check `README.md` for setup instructions
- Review error messages in terminal
- Ensure Ollama is running: `ollama serve`
- Verify model is installed: `ollama list`

---

**Happy posting! 🎉**