K-Dense-AI

K-Dense-AI / science-superpowers

Public

Composable computational-science methodology skills for AI research agents — pre-registration over TDD. A science-domain reimplementation of Superpowers.

44
1
89% credibility
Found May 29, 2026 at 105 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Shell
AI Summary

Science Superpowers is a methodology package that makes AI research assistants follow proper scientific practices. When you use an AI assistant with this package installed, it automatically guides investigations through a structured workflow: defining clear research questions, reviewing prior work, designing analysis plans, locking predictions before seeing results (pre-registration), executing reproducible analyses, investigating anomalies thoroughly, and subjecting findings to adversarial review before acceptance. The goal is to produce trustworthy, reproducible scientific findings by enforcing discipline at every step of the research process.

How It Works

1
🔬 You want to do research with AI

You have a research question or area of interest and want an AI assistant to help investigate it properly.

2
✨ Your assistant automatically knows how to help

Instead of you having to explain how to do research methodology, your AI assistant automatically loads special skills that guide the investigation the right way.

3
📝 You define your research question together

Before jumping into analysis, you and your assistant turn a fuzzy idea into a precise, testable question with clear hypotheses and what would count as an answer.

4
🔍 You plan and lock down your approach

Your assistant helps you design the analysis steps, then locks in your predictions and decision rules so you can't accidentally change them after seeing results.

5
🧪 Your assistant runs the analysis

With a clear plan locked in place, your assistant executes the analysis in a reproducible environment where everything is documented and traceable.

6
🛡️ Your work gets challenged and verified

Before you believe any result, a skeptical review kicks in to find weaknesses, and everything is verified against fresh evidence.

📚 You have trustworthy, reproducible findings

Your research is protected from common pitfalls like p-hacking, your methods are transparent, and your findings are ready to share or publish with confidence.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 105 to 44 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is science-superpowers?

Science Superpowers is a methodology framework that gives AI research agents built-in scientific discipline. Instead of letting agents jump straight into running code on data, it forces them through a structured workflow: define a falsifiable question, survey prior work, pre-register hypotheses before seeing outcomes, then execute reproducibly. It's written in Shell and works as a plugin for multiple agent harnesses including Cursor, Claude Code, Gemini CLI, and OpenCode. The core idea is applying test-driven development principles to computational science—lock your predictions before you look at the data.

Why is it gaining traction?

The hook is pre-registration. Researchers know p-hacking and HARKing (hypothesizing after results are known) are endemic problems, but most tools don't enforce safeguards. This project makes rigor automatic rather than aspirational. The skills trigger at session start, so agents don't need to be explicitly prompted to follow the methodology—they just do it. The composable skill library means you can adopt the workflow incrementally or use it wholesale.

Who should use this?

Computational researchers using AI agents for data analysis will get the most value—particularly those in fields where reproducibility and statistical rigor matter (biomedicine, social science, economics). Teams building research agent pipelines should evaluate this as a quality gate. If you're currently relying on agents that "just figure it out," this adds guardrails that prevent common methodological mistakes.

Verdict

This is a thoughtful approach to a real problem, but the 44 stars and early-stage maturity mean you should evaluate it carefully before production use. The documentation is solid and the multi-harness support is practical. With a credibility score of 0.9%, the project is credible enough to explore, but budget time for testing it against your specific workflows before trusting it with real research.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.