URL Fetching

GitAuto provides the model with a URL fetching tool that retrieves the full content of a webpage and returns it as clean markdown. The model uses this after web search to read documentation pages, API references, and changelogs that contain the details it needs.

Why This Exists

Web search returns titles and short snippets - enough to confirm that a library exists or find the right documentation page, but not enough to read the actual API surface. When the model needs to know the exact function signature, configuration format, or migration steps from one version to another, it needs to read the full page. Without this tool, the model would have to guess the details based on the snippet alone, which leads to partially correct but subtly wrong code.

How It Works

When the model calls the fetch tool with a URL, GitAuto fetches the page, strips out non-content elements (navigation, ads, footers, scripts, styles, SVGs), and finds the main content area using semantic HTML selectors like <main>, <article>, or div[role='main']. The cleaned HTML is then converted to markdown so the model can parse headings, code blocks, and links naturally.

The tool description tells the model to use it after web search and only for relevant URLs, preventing unnecessary fetches that waste time and tokens.

Related Features

• Web Search - searches the web and returns snippets to identify relevant pages
• Anti-Hallucination Prompts - system instructions that tell the model to verify before writing
• Strict Tool Schemas - ensures the fetch tool receives well-formed arguments