Google Gemini Integration Guide

Google Gemini Integration Guide

Complete guide for creating and deploying skills to Google Gemini using Skill Seekers.

Overview

Skill Seekers packages documentation into Gemini-compatible formats optimized for:

• Gemini 2.0 Flash for enhancement
• Files API for document upload
• Grounding for accurate, source-based responses

Setup

1. Install Gemini Support

# Install with Gemini dependencies
pip install skill-seekers[gemini]

# Verify installation
pip list | grep google-generativeai

2. Get Google API Key

1. Visit Google AI Studio
2. Click "Get API Key"
3. Create new API key or use existing
4. Copy the key (starts with AIza)

3. Configure API Key

# Set as environment variable (recommended)
export GOOGLE_API_KEY=AIzaSy...

# Or pass directly to commands
skill-seekers upload --target gemini --api-key AIzaSy...

Complete Workflow

Step 1: Scrape Documentation

# Use any config (scraping is platform-agnostic)
skill-seekers scrape --config configs/react.json

# Or use a unified config for multi-source
skill-seekers unified --config configs/react_unified.json

Result: output/react/ skill directory with references

Step 2: Enhance with Gemini (Optional but Recommended)

# Enhance SKILL.md using Gemini 2.0 Flash
skill-seekers enhance output/react/ --target gemini

# With API key specified
skill-seekers enhance output/react/ --target gemini --api-key AIzaSy...

What it does:

• Analyzes all reference documentation
• Extracts 5-10 best code examples
• Creates comprehensive quick reference
• Adds key concepts and usage guidance
• Generates plain markdown (no YAML frontmatter)

Time: 20-40 seconds Cost: ~$0.01-0.05 (using Gemini 2.0 Flash) Quality boost: 3/10 → 9/10

Step 3: Package for Gemini

# Create tar.gz package for Gemini
skill-seekers package output/react/ --target gemini

# Result: react-gemini.tar.gz

Package structure:

react-gemini.tar.gz/
├── system_instructions.md  # Main documentation (plain markdown)
├── references/             # Individual reference files
│   ├── getting_started.md
│   ├── hooks.md
│   ├── components.md
│   └── ...
└── gemini_metadata.json    # Platform metadata

Step 4: Upload to Gemini

# Upload to Google AI Studio
skill-seekers upload react-gemini.tar.gz --target gemini

# With API key
skill-seekers upload react-gemini.tar.gz --target gemini --api-key AIzaSy...

Output:

✅ Upload successful!
Skill ID: files/abc123xyz
URL: https://aistudio.google.com/app/files/abc123xyz
Files uploaded: 15 files

Step 5: Use in Gemini

Access your uploaded files in Google AI Studio:

1. Go to Google AI Studio
2. Navigate to Files section
3. Find your uploaded skill files
4. Use with Gemini API or AI Studio

What Makes Gemini Different?

Format: Plain Markdown (No YAML)

Claude format:

---
name: react
description: React framework
---

# React Documentation
...

Gemini format:

# React Documentation

**Description:** React framework for building user interfaces

## Quick Reference
...

No YAML frontmatter - Gemini uses plain markdown for better compatibility.

Package: tar.gz Instead of ZIP

Gemini uses .tar.gz compression for better Unix compatibility and smaller file sizes.

Upload: Files API + Grounding

Files are uploaded to Google's Files API and made available for grounding in Gemini responses.

Using Your Gemini Skill

Option 1: Google AI Studio (Web UI)

1. Go to Google AI Studio
2. Create new chat or app
3. Reference your uploaded files in prompts:

   Using the React documentation files, explain hooks
   

Option 2: Gemini API (Python)

import google.generativeai as genai

# Configure with your API key
genai.configure(api_key='AIzaSy...')

# Create model
model = genai.GenerativeModel('gemini-2.0-flash-exp')

# Use with uploaded files (automatic grounding)
response = model.generate_content(
    "How do I use React hooks?",
    # Files automatically available via grounding
)

print(response.text)

Option 3: Gemini API with File Reference

import google.generativeai as genai

# Configure
genai.configure(api_key='AIzaSy...')

# Get your uploaded file
files = genai.list_files()
react_file = next(f for f in files if 'react' in f.display_name.lower())

# Use file in generation
model = genai.GenerativeModel('gemini-2.0-flash-exp')
response = model.generate_content([
    "Explain React hooks in detail",
    react_file
])

print(response.text)

Advanced Usage

Enhance with Custom Prompt

The enhancement process can be customized by modifying the adaptor:

from skill_seekers.cli.adaptors import get_adaptor
from pathlib import Path

# Get Gemini adaptor
adaptor = get_adaptor('gemini')

# Enhance with custom parameters
success = adaptor.enhance(
    skill_dir=Path('output/react'),
    api_key='AIzaSy...'
)

Programmatic Upload

from skill_seekers.cli.adaptors import get_adaptor
from pathlib import Path

# Get adaptor
gemini = get_adaptor('gemini')

# Package skill
package_path = gemini.package(
    skill_dir=Path('output/react'),
    output_path=Path('output/react-gemini.tar.gz')
)

# Upload
result = gemini.upload(
    package_path=package_path,
    api_key='AIzaSy...'
)

if result['success']:
    print(f"✅ Uploaded to: {result['url']}")
    print(f"Skill ID: {result['skill_id']}")
else:
    print(f"❌ Upload failed: {result['message']}")

Manual Package Extraction

If you want to inspect or modify the package:

# Extract tar.gz
tar -xzf react-gemini.tar.gz -C extracted/

# View structure
tree extracted/

# Modify files if needed
nano extracted/system_instructions.md

# Re-package
tar -czf react-gemini-modified.tar.gz -C extracted .

Gemini-Specific Features

1. Grounding Support

Gemini automatically grounds responses in your uploaded documentation files, providing:

• Source attribution
• Accurate citations
• Reduced hallucination

2. Multimodal Capabilities

Gemini can process:

• Text documentation
• Code examples
• Images (if included in PDFs)
• Tables and diagrams

3. Long Context Window

Gemini 2.0 Flash supports:

• Up to 1M token context
• Entire documentation sets in single context
• Better understanding of cross-references

Troubleshooting

Issue: google-generativeai not installed

Solution:

pip install skill-seekers[gemini]

Issue: Invalid API key format

Error: API key doesn't start with AIza

Solution:

• Get new key from Google AI Studio
• Verify you're using Google API key, not GCP service account

Issue: Not a tar.gz file

Error: Wrong package format

Solution:

# Use --target gemini for tar.gz format
skill-seekers package output/react/ --target gemini

# NOT:
skill-seekers package output/react/  # Creates .zip (Claude format)

Issue: File upload failed

Possible causes:

• API key lacks permissions
• File too large (check limits)
• Network connectivity

Solution:

# Verify API key works
python3 -c "import google.generativeai as genai; genai.configure(api_key='AIza...'); print(list(genai.list_models())[:2])"

# Check file size
ls -lh react-gemini.tar.gz

# Try with verbose output
skill-seekers upload react-gemini.tar.gz --target gemini --verbose

Issue: Enhancement fails

Solution:

# Check API quota
# Visit: https://aistudio.google.com/apikey

# Try with smaller skill
skill-seekers enhance output/react/ --target gemini --max-files 5

# Use without enhancement
skill-seekers package output/react/ --target gemini
# (Skip enhancement step)

Best Practices

1. Organize Documentation

Structure your SKILL.md clearly:

• Start with overview
• Add quick reference section
• Group related concepts
• Include practical examples

2. Optimize File Count

• Combine related topics into single files
• Use clear file naming
• Keep total under 100 files for best performance

3. Test with Gemini

After upload, test with sample questions:

1. How do I get started with [topic]?
2. What are the core concepts?
3. Show me a practical example
4. What are common pitfalls?

4. Update Regularly

# Re-scrape updated documentation
skill-seekers scrape --config configs/react.json

# Re-enhance and upload
skill-seekers enhance output/react/ --target gemini
skill-seekers package output/react/ --target gemini
skill-seekers upload react-gemini.tar.gz --target gemini

Cost Estimation

Gemini 2.0 Flash pricing:

• Input: $0.075 per 1M tokens
• Output: $0.30 per 1M tokens

Typical skill enhancement:

• Input: ~50K-200K tokens (docs)
• Output: ~5K-10K tokens (enhanced SKILL.md)
• Cost: $0.01-0.05 per skill

File upload: Free (no per-file charges)

Next Steps

1. ✅ Install Gemini support: pip install skill-seekers[gemini]
2. ✅ Get API key from Google AI Studio
3. ✅ Scrape your documentation
4. ✅ Enhance with Gemini
5. ✅ Package for Gemini
6. ✅ Upload and test

Resources

• Google AI Studio
• Gemini API Documentation
• Gemini Pricing
• Multi-LLM Support Guide

Feedback

Found an issue or have suggestions? Open an issue