This wraps the Gemini CLI for handling code reviews, architectural analysis, and anything requiring massive context windows over 200k tokens. It defaults to Gemini 3 Pro, which apparently scores 76.2% on SWE-bench and handles up to 1M tokens of input. The main gotcha is approval modes: you must use yolo mode for background execution in Claude Code or it'll hang indefinitely waiting for interactive prompts you can't provide. The skill asks which model you want upfront (3 Pro for reasoning, 3 Flash for speed, or legacy 2.5 models), runs the analysis, and reminds you to start a new session for follow-ups since there's no built-in resume. Honestly most useful when you need to throw an entire codebase at an LLM and get back coherent analysis without token limit anxiety.
npx skills add https://github.com/davila7/claude-code-templates --skill gemini