LMU Munich · Universitäts-Sternwarte München
☕ Code & Coffee
June 1st 2026 · LMU Astrophysics Dept. · Presenter: Giovanni Picogna

AI Agents,
Done Right.

Introducing the lmu-usm-agent-template — the first group-wide, reproducible, citable framework for integrating AI coding agents into astrophysics research at LMU Munich.

📅
When June 1st 2026 · 2 pm
📍
Where Seminar Room South
Community draft — your input matters
First astrophysics-native agent template
ADS MCP server integration
Policy-compliant AI use in publications
GitHub Copilot Agent Mode
Reproducible by design
MNRAS · A&A · ApJ · Nature · Science
🚀
AI coding assistants are already reshaping how software is written across science. Astrophysics is no exception — and right now, every researcher is re-inventing the wheel alone. This template changes that. It encodes the group's collective knowledge — citation conventions, unit-safety rules, data-path policies, prompt-logging standards — once, so every future project starts on solid ground. This is infrastructure for the next decade of research at USM.
Main Discussion Points
01 ·

What problem does this solve?

Without a shared baseline, teams re-invent rules for safe AI use every project. Errors creep in; results become hard to reproduce. The template encodes one canonical answer, group-wide.

02 ·

Specialist agents for our domains

Five domain-specific agents: @literature-agent, @simulation-agent, @retrieval-agent, @spectral-agent, @mcmc-agent. Each understands our codes, unit conventions, and sanity checks.

03 ·

ADS via MCP — live literature in context

The NASA ADS MCP server connects directly to your coding session. Ask for references, get real BibTeX back. No hallucinated citations — ever.

04 ·

Reproducibility built-in, not bolted on

Pre-commit hooks (black, flake8, BibTeX DOI validation) + CI/CD on every PR. Prompt logs are version-controlled. Your AI-assisted analysis is fully auditable.

05 ·

Journal policies — what do they actually allow?

MNRAS, A&A, ApJ, Nature, Science each have different AI disclosure rules. This template keeps you compliant by default, with a built-in checklist for every submission.

06 ·

This is a community draft — shape it now

Version 1.0 will be out on the 1st of June, but it's only as good as the group's collective input. What's missing for your workflow? This coffee is your chance to steer the standard.

Specialist Agents in This Release
@literature-agent @simulation-agent @retrieval-agent @spectral-agent @mcmc-agent
Come Ready to Discuss
  • Are you already using GitHub Copilot or another AI agent? What works, what fails?
  • How should we cite AI assistance in our papers — and where do we draw the line?
  • Can the agent template reduce onboarding time for new PhD students and postdocs?
  • What other domain agents does the group need — RAMSES? NIRVANA? Euclid pipelines?
  • How do we prevent data leakage when working with proprietary or embargoed observations?
  • Should prompt logs become a standard supplement alongside code and data?

This is your template too. Come and change it.

All 200 refereed papers USM produces each year will be touched by AI tools within the next few years. The norms we set today will shape how the group does science tomorrow. Don't miss this conversation.

★ Star on GitHub
github.com/GiovanniPicogna/lmu-usm-agent-template
LMU Munich Astrophysics
Universitäts-Sternwarte München · June 1st 2026