What is qiaomu-heavyskill?

Qiaomu-heavyskill is a skill plugin for Claude Code that gives it a multi-agent deliberation mode for hard questions. The problem it solves is that asking a single AI a complex question yields one perspective, and you cannot easily tell whether that perspective is incomplete or missing critical blind spots. This tool instead runs 3 to 5 fully isolated AI agents in parallel, each reasoning independently about the same question without seeing the others' work, then uses OpenAI Codex to host a critical deliberation session that synthesizes all the different perspectives into a deeper conclusion. The key claim, backed by a referenced arXiv paper, is that deliberation between multiple reasoning traces is generative rather than just averaging: the synthesizer sometimes produces correct answers that none of the individual agents reached on their own. Each run produces a folder of output files including individual reasoning traces as Markdown documents, the raw Codex deliberation output, a polished final report in Markdown and HTML, and a combined PDF suitable for sharing. It supports two modes: a verification mode for questions that have objectively better answers (such as debugging a bug or checking a math derivation), and a deliberation mode for questions with no single right answer (such as choosing a software architecture or evaluating a product strategy). You trigger it through natural language prompts in Claude Code rather than specific commands. It requires both Claude Code and the OpenAI Codex plugin to be installed, since Codex serves as the deliberation moderator.