# GitAgent > An open standard for defining, versioning, and running AI agents natively in git — framework-agnostic and version-controlled. GitAgent is a git-native, framework-agnostic open standard for defining AI agents as files in a repository. It lets you version-control your agent's identity, skills, tools, memory, and compliance config using familiar git workflows like branching, PRs, and tagging. Maintained by the team at Lyzr, GitAgent exports to Claude Code, OpenAI Agents SDK, CrewAI, OpenClaw, Nanobot, Lyzr Studio, and GitHub Models from a single definition. - **Git-Native Agent Definition**: Define your agent in `agent.yaml`, `SOUL.md`, and `RULES.md` — every change is a git commit with full history, diff, and rollback support. - **Framework-Agnostic Export**: Use `gitagent export` to generate framework-specific configs for Claude Code, OpenAI, CrewAI, OpenClaw, Nanobot, Lyzr, and GitHub Models from one source of truth. - **CLI Tooling**: Install via `npm install -g gitagent`, then use `init`, `validate`, `run`, `export`, `import`, `install`, `skills`, and `audit` commands to manage the full agent lifecycle. - **SkillsFlow Workflows**: Define deterministic multi-step pipelines in YAML under `workflows/`, chaining `skill:`, `agent:`, and `tool:` steps with `depends_on` ordering and `${{ }}` template data flow. - **Live Agent Memory**: The `memory/runtime/` subfolder persists execution state across sessions via `dailylog.md`, `key-decisions.md`, and `context.md` files committed to git. - **Compliance & Governance**: First-class support for FINRA Rule 3110, SEC 17a-4, Federal Reserve SR 11-7, and CFPB frameworks; run `gitagent audit` for a full regulatory report. - **Agent Skills System**: Reusable capability modules following the Agent Skills standard, discoverable via SkillsMP marketplace, GitHub repos, or local filesystem. - **Human-in-the-Loop**: Agents can open branches and PRs for human review before merging new skills or memory updates, enabling RL-style supervised learning. - **CI/CD for Agents**: Run `gitagent validate` in GitHub Actions on every push to test agent behavior, block bad merges, and auto-deploy to production. - **Agent Registry**: Browse and share community agents at registry.gitagent.sh. ## Features - Git-native agent versioning with full commit history - Framework-agnostic export to Claude Code, OpenAI, CrewAI, OpenClaw, Nanobot, Lyzr, GitHub Models - CLI with init, validate, run, export, import, install, skills, and audit commands - SkillsFlow deterministic multi-step YAML workflows - Live agent memory via runtime/ subfolder committed to git - Compliance support for FINRA, SEC, Federal Reserve, CFPB - Reusable skills system with SkillsMP marketplace discovery - Human-in-the-loop via branch + PR review workflow - CI/CD integration via gitagent validate in GitHub Actions - Agent Registry for discovering and sharing community agents - Branch-based deployment (dev → staging → main) - Secret management via .gitignore and .env files - Agent lifecycle hooks (bootstrap.md, teardown.md) - Segregation of Duties (SOD) with conflict matrices - Knowledge tree with entity relationships and embeddings ## Integrations Claude Code, OpenAI Agents SDK, CrewAI, OpenClaw, Nanobot, Lyzr Studio, GitHub Models, GitHub Actions ## Platforms CLI, API, DEVELOPER_SDK ## Pricing Open Source ## Version 0.1.0 ## Links - Website: https://www.gitagent.sh - Documentation: https://github.com/open-gitagent/gitagent#readme - Repository: https://github.com/open-gitagent/gitagent - EveryDev.ai: https://www.everydev.ai/tools/gitagent