cosmosmesh-privacy-preservi.../AGENTS.md

1.4 KiB

CosmosMesh MVP: Architecture & Contribution

Overview

  • This workspace implements a minimal CatOpt-Bridge MVP for CosmosMesh privacy-preserving federated planning.
  • Language: Python. Package layout under src/ for a PyPI-friendly install.
  • Core MVP: LocalProblems, SharedVariables, PlanDelta, and a tiny Graph-of-Contracts (GoC) registry scaffold, plus a minimal DSL sketch.

Tech Stack

  • Python 3.x, dataclasses for lightweight data models
  • Lightweight unit tests with Python's unittest (pytest-compatible)
  • TLS-ready stubs for adapters and transport (to be wired later)

Testing & Build

  • Run tests via: bash test.sh
  • Build the package via: python3 -m build
  • This repo includes a minimal test to validate the CatOpt bridge round-trip

Contribution Rules

  • Keep changes small and focused, with clear intent and minimal surface area
  • Add tests for any new public surface
  • Do not modify public API semantics without explicit approval
  • If you add new dependencies, add them to pyproject.toml and ensure tests pass
  • Follow the style used in the repo (ASCII, concise code, docstrings where helpful)

How to extend (high level)

  • Add more adapters (e.g., rover_planner, habitat_module) under a dedicated adapters module
  • Expand the DSL sketch to cover more primitives (DualVariables, PrivacyBudget, AuditLog, PolicyBlock)
  • Implement a small simulator to exercise delta-sync and islanding scenarios