duolahypercho/fusion-fable — reverse-engineered prompt

Build me a Claude Code skill called Fusion Fable that takes one hard prompt, sends it to a small panel of models in parallel, then has Opus 4.8 compare their answers and write one stronger final answer.

I want the same prompt to go out independently, with no panelist seeing the others, then a judge step that shows consensus, contradictions, partial coverage, unique insights, and blind spots, followed by the final answer at the top. It should work with two independent Opus runs out of the box, and if Codex or agy are installed it should automatically use a richer panel with GPT 5.5 and Gemini 3.1 Pro. If those tools are missing, it should fall back cleanly.

Please make it installable with a simple script into my Claude config, add a few slash commands to force specific panel choices, and save each run to a timestamped provenance file under my Claude folder for auditing. Keep it practical and reliable for real repo work, including timeouts and safe temp work areas. Look up current Claude Code skill docs if you need to.