Agentic software engineering coordination mechanisms found in popular repositories
Methodological Harness in Agentic Software Engineering: An Empirical Study on Mining Software Repositories
Software Engineering
Summary
Coordinating software programs that act on their own, or agents, needs special tools and rules. The authors studied many software projects on GitHub to see how these coordination tools are actually used. They found that while some tools and rules are common, most projects use only parts of the full system proposed for managing these agents. Popular projects use these coordination tools more than less visible ones, but the full combined approach is rare.
What this means in practice
- •For software development teams: Identify and implement key coordination mechanisms for teams using autonomous software agents based on evidence from popular repositories.
- •For software project managers: Assess the presence and maturity of agent coordination practices to improve governance of software projects involving autonomous agents.
Authors
Jessica Díaz, Jorge Pérez, Sergio Gil-Borrás
Abstract
Context. Agentic software engineering requires mechanisms to coordinate and govern agents' work. The framework motivating this study proposes a methodological harness with eight mechanisms: context engineering, persistent shared knowledge, executable specifications, N-version mindset and parallel agents, normative specifications, structured consultation, evidence-based acceptance, and graduated autonomy. These are realized through artifacts such as rule or context files, specifications, and architectural decision records, but the framework had not been empirically validated at repository scale. Objective. We analyze to what extent and how this harness is observable in repositories with agentic activity. RQ1 characterizes adoption through artifact prevalence, breadth, co-occurrence, and temporal evolution; RQ2 examines rule files - a key observable artifact and persistent source of agent instructions - to assess how their content reflects the proposed mechanisms. Method. From the AIDev dataset of 116,211 GitHub repositories, we analyzed 5,435 using a design stratified by visibility, measured through stars as an indicator of popularity and/or reputation. For RQ1, we detected and quantified artifacts and introduction dates for the seven observable mechanisms. For RQ2, we qualitatively coded rule files from 150 repositories. Results. Population prevalence of at least one mechanism is 21.7%, versus 65.3% among the most visible repositories. Multi-mechanism configurations are rare. Where rule files exist, they almost always guide the agent and state norms, while persistent shared knowledge, executable specifications, structured consultation, and graduated autonomy appear only in a minority. Conclusions. The harness is empirically observable, but mainly through isolated mechanisms rather than the integrated system proposed by the framework.