Agentic Stacks is a package manager for domain expertise. You pull skill packs into your project and your AI coding assistant instantly gains deep knowledge about a specific domain — OpenStack deployment, Kubernetes cluster operations, Ceph storage, Docker containers, bare metal server management, and more. It's AI for ops teams — instead of searching docs and runbooks, your agent already knows the procedures, the gotchas, and the recovery steps.
Think of it like npm or pip, but instead of code libraries you're installing operational knowledge that makes your coding agent an expert operator.
A stack is a git repo containing:
When an agent reads the stack, it gains the ability to deploy, configure, troubleshoot, and upgrade the target software.
Yes. Agentic Stacks is open source (MIT license). All curated stacks in the agentic-stacks org are free. Third-party stacks are managed by their own authors.
Agentic Stacks works with any AI coding assistant that reads markdown project files. Because stacks are plain markdown, they work with any LLM tools and coding agents regardless of the underlying model:
CLAUDE.md automaticallyAGENTS.md (copy from CLAUDE.md)GEMINI.md.cursorrules or project docs.stacks/ directoryModels from OpenAI, Anthropic, Google, and open source projects can all consume stack skills since they're plain markdown.
pip install agentic-stacks
agentic-stacks init agentic-stacks/openstack-kolla my-project
cd my-project
claudeThat's it. Claude Code reads the CLAUDE.md in your project, which points to .stacks/*/CLAUDE.md. The agent now has deep expertise in whatever stacks you've pulled.
Codex CLI reads AGENTS.md. After pulling stacks:
agentic-stacks init agentic-stacks/kubernetes-talos my-cluster
cd my-cluster
cp CLAUDE.md AGENTS.md
codexThe skills are plain markdown — OpenAI models read them just as well as Claude. The routing table in CLAUDE.md/AGENTS.md tells the agent where to find each skill.
Gemini CLI reads GEMINI.md:
agentic-stacks init agentic-stacks/ceph my-storage
cd my-storage
cp CLAUDE.md GEMINI.md
geminiOpen your project directory in Cursor or Windsurf. These editors read project-level markdown files. You can either:
.stacks/*/CLAUDE.md.cursorrules or the editor's equivalentYes — that's the whole point. Compose stacks for cross-domain expertise:
agentic-stacks init agentic-stacks/openstack-kolla my-cloud
cd my-cloud
agentic-stacks pull hardware-dell
agentic-stacks pull cephYour agent now knows OpenStack deployment, Dell server management, and Ceph storage — and can combine that knowledge. "Deploy OpenStack with Ceph backend on these Dell servers" becomes a conversation, not a project.
Stack authors research official documentation, release notes, and real-world operational experience, then distill it into structured skills. Every command is verified against official docs. Every config field is checked. Version-specific behaviors are noted.
Stacks also improve over time — when operators hit issues, they feed the fix back into the stack's known issues or skill content.
Documentation tells you what a tool does. A stack teaches the agent how to operate it — the right order, the safety checks, the decision points, the gotchas, and the recovery procedures. Stacks go beyond static docs to enable real AI operations — the agent doesn't just read about a tool, it learns the judgment calls that experienced operators rely on. This takes you beyond vibe coding into structured, verifiable infrastructure operations.
Stacks include:
No. Stacks are passive knowledge — they're markdown files that teach the agent. The agent then proposes commands for you to review and approve. Your agent's existing permission model (approval gates, sandboxing, etc.) still applies.
Yes! See the authoring guide.
agentic-stacks create my-org/my-stackThis scaffolds the full structure. Write your skills, validate with agentic-stacks doctor, and publish.
Still have questions? Open an issue on GitHub.