A team of developers has launched GitAgent, an open standard that defines AI agents as files in a git repository. The project addresses the fragmentation problem in agent frameworks where switching between systems requires rewriting agent definitions. GitAgent introduces a simple three-file structure: agent.yaml for configuration, SOUL.md for personality and instructions, and SKILL.md for capabilities.

By leveraging git's native features, GitAgent provides version control for agent behavior, branching for environment promotion, and human-in-the-loop workflows through pull requests. The system supports audit trails via git blame and diff, agent forking and remixing, and CI/CD integration with GitHub Actions. The CLI tool allows direct execution of any agent repository using npx, and the spec includes optional compliance layers with regulatory mappings for financial services.

Available at https://gitagent.sh with code on GitHub, the project aims to create a portable agent definition that exports to major frameworks including Claude Code, OpenAI Agents SDK, CrewAI, Google ADK, and LangChain. The team is seeking feedback on schema design and adapter development.