Type: url
Original: https://arxiv.org/pdf/2606.27045
Status: generated
Summary
Presents the Spec Growth Engine, a framework for AI-assisted development aimed at two failure modes its author argues existing spec-driven approaches do not solve: context explosion, where an agent reasoning over a whole repository degrades as its context window fills, and silent spec-code drift, where code evolves while the specification does not and the divergence stays invisible. The framework combines a machine-readable spec graph whose nodes separate outward contract from inward design, a Spine context assembler that scopes an agent’s context to an ownership path, a hardest-first vertical-slice growth protocol with a two-layer rule, a drift gate that makes spec-code divergence a blocking merge error, and a governance model matching review to a change’s blast radius. The author presents it as a machine-enforced synthesis of established software engineering principles rather than a new paradigm, and it is a design proposal rather than an evaluated system.