What is science-superpowers?

Science Superpowers is a set of instructions and reusable workflow scripts, called skills, that you install into an AI research agent to make it follow rigorous scientific practice automatically. The idea is that when you ask an AI agent to analyze data, a default agent often jumps straight to running code. This toolkit intercepts that habit and makes the agent follow a structured process instead, without you having to ask. The workflow mirrors how careful scientists work. Before touching any data, the agent first turns your question into a precise, testable hypothesis. It then reviews what is already known about the topic, designs the analysis, and commits its predictions and decision rules to a locked document. This locking step, called pre-registration, happens before any outcomes are seen. That discipline protects against a common pitfall in data analysis where a researcher tweaks the method until the data looks good, then claims the result was the original plan. The agent also runs the analysis in a reproducible workspace with pinned software versions and fixed random seeds, investigates anomalies by root cause rather than silently discarding them, and asks a separate skeptical reviewer to look for flaws before accepting any conclusion. The toolkit has 15 individual skills, each covering one stage of the research lifecycle: framing the question, reviewing prior work, designing the analysis, pre-registering it, executing it, investigating surprises, verifying results, reviewing critically, and archiving everything. The skills trigger automatically at the right moments in a session, so the user does not need to invoke them manually. Installation requires only a POSIX shell and the user's existing agent setup. There are no third-party software dependencies. The project is a reimplementation of a software-development methodology called Superpowers, adapted for data and science work rather than code.