Home
LMU USM — AI Agent Configuration Template
A community template for using GitHub Copilot (and other AI coding agents) in reproducible, citable astrophysics research at LMU Munich / Universitäts-Sternwarte München.
Template version: 1.0 (May 2026) · Status: community draft
USM produces ~200 refereed astronomy papers per year (1,016 in 2021–2026 via NASA ADS). Primary publication venues: Astronomy & Astrophysics (~54%), MNRAS (~16%), ApJ family (~11%), Physical Review D + JCAP (~10%).
View on GitHub Use this template ▶ View Slides 📋 Pre-Meeting Poll
Presentation: AI Agents — How We Use Them, How We Cite Them
Code & Coffee, LMU Astrophysics Department, May 2026
A 41-slide reveal.js presentation covering GitHub Copilot Agent Mode, MCP & the ADS server, pitfalls in AI-assisted research, reproducibility best practices, journal disclosure policies (MNRAS, A&A, ApJ, Nature, Science), IDE landscape (Cursor, Windsurf, Google Antigravity), and agentic AI benchmarks in astrophysics (ReplicationBench, Stargazer).
What problem does this solve?
Without a shared baseline, every researcher using AI agents in their project re-invents the same rules — how to cite references safely, how to avoid uploading proprietary data, what unit conventions to use, how to log prompts for reproducibility. Errors creep in; results become hard to reproduce.
This template encodes group-level conventions once so that every Copilot / Claude / Gemini session starts from the same safe, reproducible baseline.
Research domains covered
| Domain | Key codes |
|---|---|
| Disk & planet formation | FARGO3D · PLUTO · NIRVANA-III · DustPy · RADMC-3D |
| Cosmological simulations | Magneticum · GADGET · yt · GadgetIO.jl |
| Exoplanet atmospheres | petitRADTRANS · CCF · dynesty · CARMENES · CRIRES+ · JWST |
| Large-scale structure | void statistics · SBI · Euclid pipelines |
| X-ray & galaxy clusters | XMM-Newton · Chandra · eROSITA · Sherpa · PyXSPEC |
Specialist agents
| Agent | Invocation | Purpose |
|---|---|---|
| Literature | @literature-agent |
ADS search, BibTeX retrieval |
| Simulation | @simulation-agent |
FARGO3D / PLUTO / Magneticum I/O and post-processing |
| Retrieval | @retrieval-agent |
petitRADTRANS forward model, CCF, dynesty |
| Spectral | @spectral-agent |
X-ray Sherpa / PyXSPEC fitting |
| MCMC | @mcmc-agent |
emcee / dynesty sampling, corner plots |
| Paper | @paper-agent |
LaTeX manuscript drafting, ADS citations, compile + auto-review |
See the Agents page for full documentation and example invocations.
Quick start
# 1. Create your project from this template (click "Use this template" above)
# 2. Set up the conda environment
conda env create -f envs/base.yml
conda activate lmu-astro
pre-commit install
# 3. Fill in AGENTS.md with your project context
# 4. Set your ADS API token
export ADS_API_TOKEN="your_token_here"
# Get it at: https://ui.adsabs.harvard.edu/user/settings/token
# 5. Open in VS Code — Copilot loads the baseline automatically
Safety rules built in
- References: never invented — always queried from the ADS MCP server
- Proprietary data: raw FITS / event lists are git-ignored by default
- Reproducibility: every agent task is logged in
prompts/ - CI: black · flake8 · file-size guard · BibTeX DOI check run on every commit
Agent skills
This template is designed to work with the K-Dense-AI/scientific-agent-skills skill library. Skills extend the agent’s capabilities for specific tasks (publication figures, literature reviews, statistical tests, etc.) without cluttering the baseline instructions.
Recommended skills for USM groups are documented in
.github/copilot-instructions.md §11.
Contributing
Open an issue or PR on GitHub. All USM group members are welcome to contribute agents, skills recommendations, or domain-specific AGENTS.md templates.