Show HN: GitAgent – An open standard that turns any Git repo into an AI agent
TL;DR
GitAgent defines an AI agent as three files in a git repo: agent.yaml (config), SOUL.md (personality/instructions), and SKILL.md (capabilities).
Key Points
- The format is framework-agnostic and exports directly to Claude Code, OpenAI Agents SDK, CrewAI, Google ADK, and LangChain.
- Being git-native gives you version control for agent behavior, branching for dev/staging/prod promotion, audit trails via git blame, and human-in-the-loop via pull requests.
- Agents can be forked, customized, and improved via PR like any open-source project – no framework lock-in.
Nauti's Take
The idea is elegantly simple – and that's rarely a coincidence. Anyone who has migrated an agent from one framework to another knows the pain.
Three files, git-native, framework-agnostic: it sounds almost too obvious to not already be a standard. The real test will be adoption: if OpenAI, Anthropic, and Google don't actively participate, GitAgent stays a clever hobbyist format.
But the approach of solving human-in-the-loop via pull requests is one of the most practical ideas in the current agent debate – because it reuses structures developers already understand.
Context
The real problem isn't missing AI capability but missing portability: building an agent in LangChain today means starting over in CrewAI tomorrow. GitAgent proposes an open file format to solve this fragmentation – similar to what Dockerfile did for containers. If the standard gains traction, public agent repositories could emerge where teams fork and customize agents without vendor dependencies.
That would also significantly simplify compliance and auditing in enterprise environments.
Sources