What is fusion-fable?

Fusion-Fable is a skill for Claude Code, the AI coding assistant made by Anthropic. It addresses a specific problem: when you ask a single AI model a hard question, you get one reasoning path. If you ask the same question to several models independently and then compare their answers, you often get better coverage of the problem because each model reasons differently, searches different sources, and sometimes catches what the others miss. This tool automates that process. The way it works is called a panel-then-judge pipeline. You submit a question and the skill dispatches it to two or three AI models simultaneously. Each model answers on its own without seeing the others' responses. Once all answers are collected, a final step runs using Opus 4.8 (Anthropic's most capable model as of the time this was written) to compare the responses, identify where they agreed, where they contradicted each other, what each one covered that the others missed, and where all of them may have blind spots. It then writes a final answer grounded in that structured analysis. Three panel configurations are available. The simplest runs the same prompt twice through Opus 4.8, which the README notes still produces meaningfully different answers on each run. The second pairs Opus 4.8 with GPT-5.5, accessed through OpenAI's codex command-line tool. The third adds a Gemini model as a third panelist. The skill detects which tools are installed and automatically picks the richest panel that is available. Installing it copies a set of shell scripts and configuration files into the Claude Code skills directory. After installation, you can trigger it either by asking naturally in conversation or by typing a slash command like /fusion-opus4.8 followed by your question. The trade-off is explicit: running a panel costs more in tokens and takes longer than a single answer. The README frames this as a tool for high-stakes questions where being confidently wrong carries real cost, not for routine lookups. The project is MIT licensed.