novaplan-decentralized-priv.../AGENTS.md

1.5 KiB

NovaPlan SWARM Architecture

This repository contains a minimal, testable MVP for the NovaPlan concept:

  • Decentralized, offline-friendly multi-agent planning with privacy-preserving considerations.
  • Local problem solving with federated aggregation (ADMM-like) and secure data contracts.
  • A lightweight mission ledger anchored when ground links are available.
  • Adapters for common hardware (rovers, habitat modules) and simulation-ready interfaces.

Guiding rules

  • Keep changes small and well-scoped. Favor minimal viable features that demonstrate core ideas.
  • Tests and packaging must pass before publishing. See test.sh and README for workflow.
  • All components are Python-based for this MVP unless the user explicitly requests another language.

Architecture overview

  • planner.py: Local problem and a tiny ADMM-style solver interface.
  • contracts.py: Data contracts (PlanDelta, SharedSchedule, ResourceUsage, PrivacyBudget, AuditLog).
  • ledger.py: Simple, auditable decision ledger with optional anchoring to external ground links.
  • adapters/: Lightweight stubs for rover and habitat module adapters.
  • tests/: Unit tests validating the core workflow and contract encoding/decoding.

Development workflow

  • Implement features as independent modules with small, focused tests.
  • Run test.sh to verify end-to-end viability, including packaging build check.
  • Update README.md with usage notes and API surface.

Testing commands

  • Run tests: ./test.sh
  • Build package: python -m build

Contribution

  • Open a PR with a focused feature, include tests, and ensure all tests pass.