Vibe coding relies on plain-language rule files such as AGENTS.md and CLAUDE.md to keep AI agents aligned with developer intent. A new arXiv paper, ZORO, argues that these files are the primary mechanism for human-agent alignment but remain passive: developers cannot tell at a glance whether a rule is being used or followed.
ZORO is described as an approach for adding active controls to support rule following. The abstract does not specify implementation details, but the motivation is clear: alignment should be observable rather than assumed. By making rule usage and compliance visible, ZORO aims to close the feedback gap in vibe-coding workflows.
The paper is the only source considered here, so there are no conflicting findings to compare. Its contribution is a framing shift: from static instructions to active, monitorable controls in agent alignment.