curl -sSL https://canifi.com/skills/gemini-deep-research/install.sh | bash

cp -r skills/gemini-deep-research ~/.canifi/skills/

# First, ensure canifi-env is installed:
# curl -sSL https://canifi.com/install.sh | bash

canifi-env set GOOGLE_EMAIL "your-email@gmail.com"

canifi-env set SERVICE_EMAIL "your-email"
canifi-env set SERVICE_PASSWORD "your-password"

User: "canifi, research the best productivity systems for software engineers"
Claude: Opens Gemini, initiates Deep Research query.
        Waits for completion (up to 20 min), extracts findings.
        Stores synthesized report in Notion Research Hub.

User: "canifi, research the current state of AI in healthcare"
Claude: Launches Deep Research, monitors for completion.
        Returns comprehensive analysis with key players,
        trends, challenges, and opportunities.

User: "canifi, research best practices for microservices architecture"
Claude: Initiates research, waits for completion.
        Provides detailed report on patterns, tools,
        case studies, and recommendations.

User: "canifi, compare React vs Vue vs Svelte for my next project"
Claude: Runs Deep Research on all three frameworks.
        Delivers comparison table with pros, cons,
        use cases, and recommendations.

1. Navigate to gemini.google.com
2. Click on Deep Research mode
3. Enter research query
4. Monitor progress (up to 20 minutes)
5. Extract and synthesize results
6. Store in appropriate Notion database
7. Report findings to user

// Chat input
'[aria-label="Enter a prompt here"]' or 'rich-textarea'

// Deep Research button
'[aria-label="Deep Research"]' or button with research icon

// Submit button
'[aria-label="Send message"]'

// Response container
'.response-content' or '.model-response'

// Research progress indicator
'.research-progress'

// Sources panel
'.sources-panel'

// Copy response
'[aria-label="Copy"]'

// New chat
'[aria-label="New chat"]'

// Deep Research can take up to 20 minutes
// Use periodic checking:
1. Launch research query
2. Wait 5 minutes (browser_wait_for)
3. Check for completion indicator
4. If not complete, wait another 5 minutes
5. Repeat until complete or 20 minute timeout
6. Extract results when done