Science with large language models.
The first open collection of tested workflows, prompts and skills for science.
Background: Hubble XDF, NASA/ESA
Large language models are changing the way research is done, and fast. Knowing how to use them well is now part of a researcher’s job.
This is a place to share what works: workflows, skills, prompts and agent configurations for modern language models, each tested on real research. Anyone can contribute. If you use material from this site, please cite it: that is how the people who shared it get credit for their work.
What you’ll find here
- Workflows
- Step-by-step procedures for a research task, with the exact prompts and the checks to run on the result.
- Prompts
- Single prompts that worked on real research tasks, with an example of the output and their known limits.
- Skills
- Packaged instructions and scripts that an AI agent loads when a task needs them, ready to install.
- Agent configs
- CLAUDE.md and AGENTS.md templates that teach coding agents the conventions of a research project.
- Model checks
- Small, repeatable tests of models on real scientific tasks, with dated results.
- Reviews
- A second, AI-assisted peer review of published papers, with the model, the prompt and the date, checked by a scientist.
Workflows by research phase
- LiteratureSearch, read and synthesise papers with ADS and arXiv
- ProposalsObserving and funding proposals, within the rules
- Data and codePipelines, legacy code, CASA and FITS handling
- AnalysisStatistics, fits, units and derivations
- WritingDrafts, figures and language review
- ReviewPre-submission checks and referee replies
Latest workflows
- Porting legacy IDL code to Python
Translate an IDL routine with a coding agent and prove the Python version gives the same numbers.
- Literature search with NASA ADS and a language model
Build the query yourself, export abstracts from ADS, and let the model map the field with every claim tied to a bibcode you can check.