mmthebest/research-architect

Research Architect is a skill suite for AI coding agents (Claude Code and Codex) that guides researchers through the process of turning a rough topic into a structured academic paper draft. The README is written in both Chinese and English. The tool is installed into the agent's skills directory, after which you trigger the workflow by typing "research-architect" in your agent session. The workflow runs in a defined sequence: start with a raw idea, map it against existing literature, narrow down to a specific research question and gap, build a research spine, design the study, plan experiments and analysis, collect and organize evidence, register claims, gather citation support for each claim, write a section blueprint, produce a first draft, and finally run an audit that generates a revision queue. The output is a folder of structured documents at paper_output/ covering each stage of that trail. A central feature is how the tool treats reference papers. Instead of treating a cited paper as a source of facts to quote, Research Architect analyzes published papers as examples of research design: how they framed the problem, how they controlled the scope, how they defined the gap, how they arranged baselines and controls, how they organized figures, and how they kept each claim within what the evidence actually supports. The goal is to apply that same structural logic to your own work. The problems the tool targets are common in academic writing: a topic that stays too broad, a contribution that sounds significant but lacks an evidence chain, claims that are stronger than the study design justifies, figures that display results without carrying an argument, and references that appear in the introduction but do not actually shape the experimental decisions. The project is MIT licensed and installable from either the repository's dist folder or a provided release tarball.