community
/
catopt-graph-graph-calculus-driven-compo
17
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
catopt-graph-graph-calculus.../AGENTS.md

2.1 KiB
Raw Blame History

CATOPT-GRAPH AGENTS

Overview

  • This repository contains a minimal, working MVP of CatOpt-Graph: a graph-calculus-inspired orchestration layer for compositional optimization across edge devices.
  • The MVP focuses on a simple asynchronous-like ADMM-lite solver, contract primitives (Objects, Morphisms, Functors), and two starter adapters (rover and habitat) that map to canonical representations.

Architecture (high-level)

  • core: Objects, Morphisms, Functors, and a tiny contract registry skeleton.
  • admm_lite: a lightweight, fault-tolerant, delta-sync solver for two agents solving a simple distributed objective with a shared constraint.
  • adapters/: two starter adapters (rover and habitat) provide readState/exposeLocalProblemData/applyCommand interfaces.
  • transport: placeholder TLS/REST-like transport surface (mocked in MVP).
  • governance: lightweight auditing placeholder.
  • tests: unit tests for the ADMM-lite core.

How to run tests

  • Ensure Python 3.10+ is installed.
  • Run: bash test.sh

Development workflow

  • Use the provided DAG of modules to extend adapters and add new ones.
  • All changes should be backed by tests.

Note

  • This is a minimal, opinionated MVP to bootstrap cross-domain interoperability. It is not a full production system.

What we're adding now (MVP roadmap refinements):

  • Core ontology extension: versioned ContractRegistry (contracts per name/version) for interop with adapters.
  • Bridge layer: a lightweight to_canonical / from_canonical mapping to connect domain LocalProblem data to the CatOpt canonical form.
  • Adapters: two starter adapters (rover, habitat) already present; bridge will enable canonical data exchange.
  • ADMM-lite core: keep the existing two-agent toy solver as a testbed; extend with delta-sync semantics conceptually in code comments and tests.
  • Governance / contracts: hook up a minimal conformance story via the contract registry + bridge; add a test to verify registry behavior.
  • MVP testing: unit tests for contract registry, end-to-end tests for bridge mapping (stubbed inputs/outputs) and existing admm_lite tests for solver stability.
  • Documentation: update README to reflect the MVP extension and how to run tests.