Skip to content

Governance

This document describes the governance model for the Agent Governance Toolkit (AGT).

Project Scope

The Agent Governance Toolkit provides runtime governance for autonomous AI agents: deterministic policy enforcement, zero-trust identity, execution sandboxing, and reliability engineering.

Roles

Maintainers

Maintainers have merge authority and are responsible for the project's technical direction, security posture, and release management. See MAINTAINERS.md for the current list.

Responsibilities: - Review and merge pull requests - Triage security vulnerabilities (MSRC coordination) - Manage releases and signing (ESRP) - Enforce contribution policies

Contributors

Anyone who submits a pull request, files an issue, or participates in discussions. See CONTRIBUTING.md for guidelines.

Community Extension Authors

External contributors who build integrations under agent-governance-python/agentmesh-integrations/. Extensions are community-maintained and clearly separated from core.

Core vs Community Extension Boundary

Path Ownership Review Policy
agent-governance-python/agent-os/ Microsoft maintainers Maintainer approval required
agent-governance-python/agent-mesh/src/ Microsoft maintainers Maintainer approval required
agent-governance-python/agent-hypervisor/ Microsoft maintainers Maintainer approval required
agent-governance-python/agent-sre/ Microsoft maintainers Maintainer approval required
agent-governance-python/agent-compliance/ Microsoft maintainers Maintainer approval required
agent-governance-python/agent-runtime/ Microsoft maintainers Maintainer approval required
agent-governance-python/agent-marketplace/ Microsoft maintainers Maintainer approval required
agent-governance-dotnet/ Microsoft maintainers Maintainer approval required
agent-governance-python/agentmesh-integrations/ Community + maintainers Maintainer review, community may author
docs/integrations/ Community + maintainers Maintainer review, community may author
docs/adr/ Community + maintainers Maintainer review for proposed ADRs
examples/ Community + maintainers Maintainer review

Core packages (agent-os, agent-mesh, agent-hypervisor, agent-sre, agent-compliance, agent-runtime, agent-marketplace, agent-governance-dotnet) are maintained exclusively by Microsoft. External contributions to core require a prior discussion in a GitHub Issue and explicit maintainer approval before a PR is opened.

Community extensions under agent-governance-python/agentmesh-integrations/ are welcome from any contributor. Extensions must not modify core packages. Each extension must include its own README, tests, and license notice.

Decision Making

  • Technical decisions are made by maintainers via GitHub Issues and ADRs (Architecture Decision Records) in docs/adr/.
  • Security decisions follow the SECURITY.md process and coordinate with MSRC when applicable.
  • Roadmap priorities are set by the maintainer team with community input via GitHub Discussions and Issues.

Releases

  • Releases follow Semantic Versioning.
  • All packages are signed via ESRP (Microsoft's approved signing service).
  • Python packages are published to PyPI, npm packages to npmjs.com, NuGet packages to NuGet.org, Rust crates to crates.io.
  • Release notes are published in RELEASE_NOTES_*.md and GitHub Releases.

Code of Conduct

This project follows the Microsoft Open Source Code of Conduct.

License

MIT License. See LICENSE.

Amendments

This governance document may be amended by maintainer consensus. Changes are tracked via pull requests to this file.