What’s new here

This repo holds only a README with the abstract and citation for a preprint (arXiv:2608.25512) that works out the theory behind Cordis, a meta-framework for dynamic component systems. The paper gives formal definitions for revertible effects (side effects that can be undone on component removal) and reactive coeffects (dependencies that drive component activation and deactivation), then proves spatiotemporal composability holds for whole systems of interleaved components.

What it does

The paper identifies two problems in dynamic composition: temporal composability (cleanly reverting a component’s effects when it is removed) and spatial composability (declaratively managing inter-component dependencies). It lifts classical effect and coeffect theory to runtime mechanisms, defines a unified context type that mediates both, and proves an observational equivalence result showing components can interleave without disturbing each other.

The implementation target is Cordis: a meta-framework with a core library for effect tracking and coeffect resolution, plus a declarative component loader that handles configuration reconciliation and hot module replacement.

Who it’s for

PL researchers and framework authors who want the formal foundations behind plugin systems or self-evolving agent harnesses. Also useful for engineers building component loaders or hot-reload systems who want to understand the formal guarantees the paper establishes for dynamic composition.

Try it

There is no code in this repository. Read the paper at the link below.

https://arxiv.org/abs/2608.25512

The arXiv page always serves the latest revision. The BibTeX key is shi2026cordis.

How mature is it

3,009 stars and 141 forks. The repo has one contributor, was created 2026-08-13, last pushed 2026-08-28, and has 5 commits in the last 90 days. No releases, no license. It is explicitly described as a preprint under active revision.