Developer sivasurend has released GitAgent, an open standard that addresses framework lock-in in AI agent development by storing agents as version-controlled files in Git repositories. The specification enables developers to build agents once and deploy across multiple frameworks including Claude Code, OpenAI Agents SDK, CrewAI, and others.
GitAgent Stores Agents as Plain Files in Git Repositories
The GitAgent specification structures AI agents as plain files committed to Git repositories. Each agent consists of agent.yaml for configuration, SOUL.md for identity and personality definitions, SKILL.md for capabilities, and additional artifacts for knowledge, memory, and compliance. This file-based approach makes agents portable across frameworks and enables standard version control workflows.
Version Control Delivers Free Agent Management Features
By leveraging Git as the storage layer, GitAgent provides several capabilities without additional tooling:
- Version control for agent behavior enables rollback of problematic prompts like reverting bad commits
- Branch-based environment promotion supports dev → staging → production workflows
- Pull request workflows enable human-in-the-loop reviews before merging agent improvements
- Git blame and diff provide complete audit trails of agent evolution
- Standard forking workflows allow developers to remix public agents and contribute improvements upstream
- CI/CD integration through GitAgent validate in GitHub Actions
Framework-Agnostic Runtime Exports
GitAgent exports to multiple runtime environments without code changes. Supported frameworks include Claude Code, OpenAI Agents SDK, CrewAI, Lyzr, OpenClaw, Nanobot, and raw system prompts. Developers install via npm install -g gitagent and run agents with npx @open-gitagent/gitagent run -r https://github.com/user/agent -a claude.
Optional Compliance Layer for Regulated Industries
The specification includes an optional compliance framework with risk tier definitions, regulatory mappings for FINRA, SEC, and SR 11-7 requirements, and audit report generation via GitAgent audit. This addresses deployment requirements in regulated sectors.
Key Takeaways
- GitAgent stores AI agents as version-controlled files in Git repositories, eliminating framework lock-in
- The standard exports to multiple runtimes including Claude Code, OpenAI Agents SDK, CrewAI, and others
- Git-native architecture provides built-in version control, branching, pull request workflows, and audit trails
- Installation requires npm install -g gitagent with execution via npx commands
- The specification is MIT-licensed with community-driven development at https://gitagent.sh