Hey HN,
We built GitAgent — a git-native, framework-agnostic open standard
for
defining AI agents. The idea is simple: your repo IS the agent.
npx @open-gitagent/gitagent@latest run -r
https://github.com/shreyas-lyzr/architect -a claude
That one command clones the repo, reads agent.yaml + SOUL.md +
skills/,
and launches it with Claude Code. Switch `-a openai` or `-a lyzr`
to run
the same agent on a different framework. No code changes.
Goal: is keep the gitagent as a open standard ledger for an AI Agent in a VM, with versioned memories, skills etc.
An agent is just files:
my-agent/
├── agent.yaml # config: model, skills, tools
├── SOUL.md # identity, personality, expertise
├── RULES.md # boundaries, constraints
├── skills/ # capabilities
├── knowledge/ # reference docs
└── memory/ # persistent context
Why Git? Because agents need the same things code needs —
versioning,
branching, diffing, code review, and collaboration. We already have
the best tool for that.
What we've built:
- CLI that validates, exports, imports, and runs agents
- 8 adapters: Claude Code, OpenAI, CrewAI, OpenClaw, Nanobot, Lyzr,
GitHub Models
- Public registry at registry.gitagent.sh — browse and share agents
- Compliance framework (FINRA, SEC, Fed) — audit logging,
human-in-the-loop, kill switches
Try it: https://gitagent.sh
Registry: https://registry.gitagent.sh
GitHub: https://github.com/open-gitagent/gitagent
Discord: https://discord.gg/hVZV8Xyjdc
Patterns
1. Human-in-the-Loop — Agents create a branch and PR for human review before updating memory or skills.
2. Live Agent Memory — memory/runtime/ stores dailylog.md, key-decisions.md, and context.md to persist state across sessions.
3. Agent Versioning — Every agent change is a git commit. Roll back broken prompts or revert bad skills anytime.
4. Shared Context via Monorepo — Root-level context.md, skills/, tools/, and knowledge/ are inherited by all agents automatically.
5. Branch-based Deployment — Promote agent changes through dev → staging → main, just like shipping software.
6. Knowledge Tree — knowledge/ folder stores entity relationships as a directory tree with embeddings for structured reasoning.
7. Agent Forking & Remixing — Fork a public agent repo, customize SOUL.md, add skills, and PR improvements back upstream.
CI/CD for Agents — Run gitagent validate on every push via GitHub Actions. Test, block bad merges, auto-deploy.
8. Agent Diff & Audit Trail — git diff shows what changed; git blame traces every line to who wrote it and when.
Tagged Releases — Tag stable versions like v1.1.0. Pin production to a tag, canary on staging, roll back instantly.
9. Secret Management — Agent config is version-controlled; API keys stay in .env excluded via .gitignore.
10. Lifecycle Hooks — hooks/bootstrap.md and hooks/teardown.md define what an agent does on startup and shutdown.
11. Git-Native — Version control, branching, diffing, and collaboration are built in.
12. Framework-Agnostic — Works with Claude Code, OpenAI, CrewAI, LangChain, and more.
13. Compliance-Ready — First-class FINRA, SEC, and Federal Reserve support with audit logging and supervision.
14. Composable — Skills, tools, sub-agents, and workflows can extend, depend on, and delegate to each other.
MIT licensed. Would love feedback.
