gemini

name	gemini
description	All-purpose Gemini 3 Pro client with Thinking enabled. Query, analyze files (MP4, PDF, images), YouTube videos, generate/edit images. Uses browser cookies - no API key required.

<gemini_skill> Gemini 3 Pro Assistant <primary_goal>Provide access to Gemini 3 Pro with Thinking for queries, file analysis, and image generation</primary_goal>

A comprehensive Gemini 3 Pro skill using gemini-webapi for cookie-based authentication. Supports text queries, file analysis (video, PDF, images), YouTube analysis, and image generation/editing. Uses "Thinking with 3 Pro" mode by default for enhanced reasoning. General questions with Thinking enabled for complex reasoning Upload MP4 files for summarization, timestamp finding, content extraction Pass YouTube URLs for video understanding via extension Upload PDFs and documents for summarization and Q&A Upload images for description, OCR, analysis Create images from text prompts Modify existing images with natural language Automatic web search for current information View model's reasoning process with --show-thoughts Chrome browser logged into gemini.google.com First time setup: cd ~/.claude/skills/gemini && python3 -m venv .venv && source .venv/bin/activate && pip install -r requirements.txt On macOS, allow Keychain access when prompted (first run) Text query: ```bash ~/.claude/skills/gemini/webapi "Explain quantum computing" ```

Analyze video file:
```bash
~/.claude/skills/gemini/webapi "Summarize this video" --file video.mp4
```

Analyze YouTube video:
```bash
~/.claude/skills/gemini/webapi "What are the key points?" --youtube "https://youtube.com/watch?v=..."
```

Analyze document:
```bash
~/.claude/skills/gemini/webapi "Summarize this report" --file report.pdf
```

Generate image:
```bash
~/.claude/skills/gemini/webapi "A sunset over mountains" --generate-image sunset.png
```

Edit image:
```bash
~/.claude/skills/gemini/webapi "Make the sky purple" --edit photo.jpg --output edited.png
```

Show thinking process:
```bash
~/.claude/skills/gemini/webapi "Solve step by step: What is 15% of 240?" --show-thoughts
```

<cli_options> | Option | Description | |--------|-------------| | --file, -f FILE | Input file (MP4, PDF, PNG, JPG, etc.) | | --youtube URL | YouTube video URL to analyze | | --generate-image FILE | Generate image and save to FILE | | --edit IMAGE | Edit existing image (use with --output) | | --output, -o FILE | Output file path for image operations | | --aspect RATIO | Aspect ratio for images (16:9, 1:1, 4:3, 3:4) | | --show-thoughts | Display model's thinking process | | --model MODEL | Model to use (default: gemini-3.0-pro) | | --json | Output response as JSON | | --help, -h | Show help | </cli_options>

1. User requests analysis, generation, or query 2. Determine appropriate mode (query, file analysis, image gen) 3. Construct command with appropriate flags 4. Run webapi script 5. Process and present results ```bash ~/.claude/skills/gemini/webapi "What are the implications of quantum computing for cryptography? Think through this carefully." --show-thoughts ```

<example name="Video meeting summary">
  ```bash
  ~/.claude/skills/gemini/webapi "Create meeting notes: attendees, topics, decisions, action items" --file meeting.mp4
  ```
</example>

<example name="YouTube tutorial extraction">
  ```bash
  ~/.claude/skills/gemini/webapi "List all steps with timestamps" --youtube "https://youtube.com/watch?v=VIDEO_ID"
  ```
</example>

<example name="Current events (grounded)">
  ```bash
  ~/.claude/skills/gemini/webapi "What are the latest developments in AI this week? Search the web."
  ```
</example>

<example name="Image generation with aspect ratio">
  ```bash
  ~/.claude/skills/gemini/webapi "A cyberpunk cityscape at night" --generate-image city.png --aspect 16:9
  ```
</example>

<example name="Document analysis">
  ```bash
  ~/.claude/skills/gemini/webapi "Extract the key findings and recommendations" --file research_paper.pdf
  ```
</example>

Cookie or initialization errors Ensure you're logged into gemini.google.com in Chrome. On macOS, allow Keychain access.

<issue name="No images generated">
  <symptom>Script runs but no image output</symptom>
  <solution>Rephrase prompt. Some content may be filtered. Try being more explicit.</solution>
</issue>

<issue name="Module not found">
  <symptom>ImportError for gemini_webapi</symptom>
  <solution>Run: cd ~/.claude/skills/gemini && source .venv/bin/activate && pip install -r requirements.txt</solution>
</issue>

<issue name="YouTube not working">
  <symptom>Cannot analyze YouTube video</symptom>
  <solution>Ensure YouTube extension is enabled in your Gemini web settings at gemini.google.com</solution>
</issue>

<issue name="No thinking shown">
  <symptom>--show-thoughts returns nothing</symptom>
  <solution>Thinking may be skipped for trivial queries. Try a more complex question, or the model may have used code execution instead.</solution>
</issue>

<important_notes> Always uses "Thinking with 3 Pro" mode for enhanced reasoning Google Search grounding is automatic - just ask about current events YouTube extension must be enabled in gemini.google.com settings Cookie-based auth - no API key needed, uses your Google account quota First run on macOS prompts for Keychain access </important_notes> </gemini_skill>