Skip to content

AGENTS.md

Latest commit

 

History

History
53 lines (39 loc) · 2.59 KB

AGENTS.md

File metadata and controls

53 lines (39 loc) · 2.59 KB

AGENTS.md

This file provides guidance to agents when working with code in this repository.

Repository Overview

This repository contains a Docker-based cosmological data analysis environment designed for cosmic shear, CMB lensing, and cross-correlation analyses. The container provides a comprehensive scientific computing stack without conda, built on Python 3.12 with specialized astronomy and cosmology libraries.

Development Workflow

Making Changes

The workflow is straightforward:

  1. Modify repository contents (typically Dockerfile or scripts/)
  2. Commit changes
  3. Push to trigger automated build

Container Build Process

  • Build Location: GitHub Actions only - no local builds or testing
  • Triggers: Pushes to main/master/develop branches, tags, and manual dispatch
  • Pipeline: Build → Smoke tests → Push to GitHub Container Registry (ghcr.io/cailmdaley/containers)
  • Testing: Automated validation via scripts/test.sh during CI

Architecture and Environment

Container Design

  • Base: Python 3.12 on Debian Bookworm
  • Build Strategy: Single-stage image consolidating ShapePipe dependencies with agent-friendly CLI tools
  • Entry Point: Bash shell with working directory at /workspace

Scientific Computing Stack

The container is organized around three primary domains:

  1. Astronomy Libraries: Core astronomical data handling (astropy, regions, reproject, skyproj)
  2. Observational Cosmology: Specialized tools for cosmic shear analysis (treecorr, pyccl, clmm, ngmix) and CMB analysis (lenspyx, ducc0, camb)
  3. Data Analysis: Standard scientific Python stack with performance libraries (numpy, numba, scipy, pandas)

Developer Tools Integration

The container includes modern CLI tools optimized for agent workflows:

  • Search/Navigation: ripgrep, fd, fzf, bat
  • Data Tools: jq, yq, sqlite3, httpie
  • System Monitoring: btop, htop, hyperfine
  • Text Processing: pandoc, quarto for scientific writing

Dependency Management

  • Git-based packages: Several packages installed directly from Git repositories for latest features
  • No conda: Pure pip-based installation for faster builds and simpler dependency resolution
  • Relaxed pins: Dependencies use minimal version constraints for rapid development

Key Files

  • Dockerfile: Main container definition with all dependencies and tools
  • scripts/test.sh: Smoke tests that validate core library imports and tool versions
  • .github/workflows/deploy-image.yml: CI pipeline for automated building and publishing