Software development is a collaborative activity, where multiple participants contribute to the evolution of a codebase. This collaborative nature is what makes governance key. Especially in Open-Source Software (OSS) development, where due to its open nature, OSS projects often face challenges in decision-making and management. One example is the participant inequality problem: everybody uses OSS, but very few contribute back. The lack of key governance information deters potential contributors, as they might fear hidden power relations or do not know how to effectively contribute.

On top of that, the software development landscape is undergoing a profound transformation. Projects now involve not only diverse human contributors but also increasingly autonomous AI agents. While this human-agent collaboration is promising, it also exposes a critical gap: the lack of governance frameworks designed to explicitly and holistically manage such multifaceted participation.

In our vision paper “Towards Automated Governance: A DSL for Human-Agent Collaboration in Software Projects”, accepted at the 40th International Conference on Automated Software Engineering (ASE’25), we address this gap by proposing the foundational concepts of a Domain-Specific Language (DSL) to define and enforce governance policies with a particular focus on human and agent diversity. This work is related with BESSER and MOSAICO projects.

We invite you to share your thoughts on governance in OSS and human–agent collaboration to help us better understand the community needs. Take our short survey here!

Our Proposal

We designed a DSL, coupled with its runtime decision engine, to define and enforce governance policies in software projects. Our DSL adapts to the current reality of software development involving participants with different profiles, the possibility of having uncertain AI agents, and decision procedures with different participant weighting systems. The architecture of our proposal can be interpreted as:

The users can write their governance policies in a file, which will be later read by the engine and enforce them in the specified platform (for now only on GitHub). The policy file contains the governance conforming to our grammar, which is defined with the following metamodel:

The metamodel is aimed at representing four main aspects: (1) scope (i.e., where the policy is applied); (2) participants (i.e., who is involved in the
policy), (3) policies (i.e., what the policy is about), and (4) conditions (i.e., when it should be applied). Governance models conforming to this metamodel represent specific sets of governance policies (cf. Policy model).

We defined our DSL as a textual language following a block-based structure. As an example, we show below a supermajority voting policy for changes in the Chinese documentation. As such, contributors that speak Chinese have a higher voting weight. Agents can also contribute, and their voting weight is based on the confidence they have on the results. Additionally, changes must be approved within 7 days.

We implemented the runtime decision engine using the BESSER Agentic Framework. The engine captures events of associated platforms (for now we have integration with GitHub), applies policies based on scope (e.g., a pull request), and resolves decisions automatically while at the same time maintaining a Runtime model for traceability purposes.

Future Steps

This is the first step towards a more diverse, agentic and transparent collaboration in software development. The next steps we will focus on are:

• Usability: by offering alternative UIs, like chatbots or a form-based syntax, to reduce overhead of learning a new syntax.
• Modularity: by addressing multiple platforms (e.g., GitLab) and use-cases.
• Impact: Does our governance layer enable more complex multi-agent systems to collaborate?
• Beyond software: we plan to explore other domains that could benefit from explicit governance policies.

Happy to discuss with you about any of these directions, and share your thoughts in our survey!