Creating Wiki Pages with AI Agent

You can leverage the Antigravity App powered by Gemini models to automatically research, draft, and format wiki pages. By providing the AI agent access to all related repositories, it can cross-reference code, graphics, and audio to create accurate and comprehensive documentation.

ℹ️ Note: You may need a paid Google One AI subscription to use recommended AI agents. You can subscribe via the Google One AI plan subscription page. The recommended AI agent for most tasks is Gemini 3.5 Flash (Medium), which may require the $100 Google AI Ultra plan due to high token usage. If you have only the Google AI Pro plan, the recommended agent is Gemini 3.1 Pro (High) (because it consumes fewer tokens), except for seldomly using Gemini 3.5 Flash (Medium) for hard tasks.

📂 Local Repository Setup

Ensure you have the following repositories cloned locally on your computer:

• GnollHackWiki (The wiki content file repository, for creating and editing pages)
• GnollHack (For researching game mechanics, stats, and source code)
• GnollHackTileSet (For finding sprite assets)
• GnollHackSoundSet (For referencing sound effects and voiceovers)

⚙️ Antigravity Workspace Configuration

1. Open the Antigravity App.
2. Ensure that your workspace is configured so the agent can access all four repository folders (GnollHackWiki, GnollHack, GnollHackTileSet, and GnollHackSoundSet). This allows the agent to read and search across the entire project ecosystem.
3. The agent will automatically discover the .agents/skills/wiki_editing skill, ensuring it follows the repository's strict formatting, linking, and image guidelines.

🧠 Model Selection and Reasoning Levels

The Antigravity App allows you to select between different models and reasoning levels. You should choose the appropriate combination based on the complexity of the wiki page you are creating:

🏆 Gemini 3.5 Flash (Medium)

This is the recommended model for most tasks. It utilizes a moderate compute budget for balanced reasoning, self-correction, and exploration, making it highly effective for standard wiki creation and editing tasks. It is also required when image conversion to WebP is needed (or other advanced tool use). To use this model for most tasks, you may need the $100 Google AI Ultra plan due to high token usage.

🛠️ Gemini 3.1 Pro (High)

This model is recommended if you have only the Google AI Pro plan, except you may seldomly use Gemini 3.5 Flash (Medium) for hard tasks. Gemini 3.1 Pro (High) utilizes a large compute budget for extensive chain-of-thought, self-correction, and exploring multiple paths before generating the final output. It is highly capable of complex tasks where the AI needs to piece together information from multiple sources. However, Gemini 3.5 Flash (Medium) works better with Antigravity than Gemini 3.1 Pro (High).

⚡ Gemini 3.5 Flash (Low)

This model is recommended for simple bulk operations. It generates output with minimal internal chain-of-thought, making it the fastest option for straightforward tasks such as formatting existing text, fixing typos, or creating simple pages.

🖼️ Image Conversion with FFmpeg

The wiki requires all images to be in WebP format. Both Gemini 3.5 Flash reasoning levels (Low and Medium) can automatically convert image assets to WebP using terminal commands if FFmpeg is installed on your computer.

If you do not have it installed, you can download FFmpeg here and ensure it is added to your system's PATH.

🤖 Common AI Tasks

📝 Checking Grammar and Spelling

• Recommended Model: Gemini 3.5 Flash (Low)
• Reasoning: Simple, straightforward task that requires minimal reasoning and is very fast.

💡 Example Prompt: Check the grammar and spelling in the "Alchemy.md" page and fix any typos. Do not change the overall tone or structure of the document.

📊 Formatting Text and Tables

• Recommended Model: Gemini 3.5 Flash (Low)
• Reasoning: Formatting existing data into tables or lists is a quick, bulk operation that doesn't need deep reasoning.

💡 Example Prompt: Format the monster stats list in "Monsters.md" into a Markdown table with columns for Name, Base Level, Health, and Armor Class.

🔍 Researching and Creating a New Page

• Recommended Model:
  • Gemini 3.5 Flash (Medium) (with Google AI Ultra plan)
  • Gemini 3.1 Pro (High) (with Google AI Pro plan)
• Reasoning: Complex task requiring searching code, synthesizing information from multiple repositories, and generating new formatted content. May require WebP conversion.

💡 Example Prompt: Research the 'Archon' monster in the GnollHack codebase. Create a new wiki page for it under /Monsters/, following the wiki conventions. Find its image in the GnollHackTileSet repository, ensure it is in WebP format, and embed it correctly.

⚖️ Verifying Wiki Information Against Codebase

• Recommended Model:
  • Gemini 3.5 Flash (Medium) (with Google AI Ultra plan)
  • Gemini 3.1 Pro (High) (with Google AI Pro plan)
• Reasoning: Requires comparing human-written text with C source code data structures and resolving any discrepancies.

💡 Example Prompt: Verify the damage stats of the "Vorpal Blade" in "Artifacts.md" against the C source code in the GnollHack repository. If there are any discrepancies, update the wiki page to match the code.