agentic-stacks 0.1.0

Installable skill packs that give AI agents deep domain expertise.

Agentic Stacks

Installable skill packs that give AI agents deep domain expertise. A stack is a git repo that teaches an agent how to operate in a specific domain — deploying OpenStack, bootstrapping Kubernetes, managing server hardware, and more.

Pull a stack into your project and your AI agent instantly knows how to deploy, manage, troubleshoot, and upgrade the target software. Stacks teach agents and humans — amplifying what you can accomplish together.

How It Works

# Start a project using a stack
agentic-stacks init agentic-stacks/openstack-kolla my-cloud
cd my-cloud

# Now talk to the agent — it knows OpenStack
# "I need an HA OpenStack cloud with OVN networking and Ceph storage
#  on these 10 hosts..."

The agent reads the stack's skills, asks the right questions, and creates the deployment configs. Everything it creates goes into your repo — reproducible, version-controlled, yours.

Compose multiple domains

Need hardware expertise alongside your platform stack? Pull in more skills:

agentic-stacks pull hardware-dell    # now it knows Dell servers too
agentic-stacks list                  # see what's loaded

The agent reads all stacks and combines their expertise — hardware provisioning, platform deployment, and everything in between.

Capture learnings as you go

Hit an issue? Ask your agent to document it. Stacks get smarter over time — every workaround, gotcha, and fix feeds back into the stack for the next person.

"That NTP fix we just did — add it to known issues for this version."

Available Stacks

Stack	Target	Skills
docker	docker	21
kubernetes-talos	talos-linux	20
openstack-kolla	openstack	8
ceph	ceph	17
hardware-dell	Dell PowerEdge	18
hardware-hpe	hpe-ilo	16
hardware-supermicro	Supermicro BMC	17
frr	frr	35
ipxe	ipxe	20

Browse all stacks at agentic-stacks.com/stacks.

Install

pip install agentic-stacks

CLI

# Start a new project from a stack
agentic-stacks init agentic-stacks/openstack-kolla my-project

# Pull stacks into .stacks/
agentic-stacks pull                    # pulls all from stacks.lock
agentic-stacks pull hardware-dell      # add another stack

# Manage stacks
agentic-stacks list                    # see loaded stacks
agentic-stacks update                  # update all to latest
agentic-stacks update --check          # check without updating
agentic-stacks remove hardware-dell    # remove a stack

# Search for stacks
agentic-stacks search kubernetes

# Create a new stack (for stack authors)
agentic-stacks create my-org/my-stack

# Validate a stack
agentic-stacks doctor --path ./my-stack

What a Stack Looks Like

A stack is a git repo with this structure:

openstack-kolla/
├── CLAUDE.md               # Agent entry point — the expertise guide
├── stack.yaml              # Manifest — identity, skills, metadata
├── skills/                 # Markdown knowledge — teaches the agent
│   ├── deploy/
│   ├── health-check/
│   ├── config-build/
│   └── diagnose/
└── ...

The CLAUDE.md is the product — it's what makes the agent an expert. The skills directory contains detailed knowledge the agent references during operations.

What a User's Project Looks Like

After init and working with the agent:

my-cloud/
├── .stacks/                # pulled stack repos (gitignored)
│   ├── openstack-kolla/    # platform expertise
│   └── hardware-dell/      # hardware expertise
├── CLAUDE.md               # points agent to .stacks/*/CLAUDE.md
├── stacks.lock             # pinned stack references
├── globals.yml             # kolla-ansible config (agent created this)
├── inventory/hosts.yml     # ansible inventory (agent helped build this)
└── ...

The output is native format for whatever tool the stack wraps. No custom formats — just the configs the tool expects.

Distribution

Stacks are git repos. Pull clones them. No package managers, no tarballs.

• Curated stacks live under the agentic-stacks GitHub org
• Third-party stacks live in their own repos — pull by org/name

agentic-stacks pull openstack-kolla           # → github.com/agentic-stacks/openstack-kolla
agentic-stacks pull someuser/their-stack      # → github.com/someuser/their-stack

Author a Stack

See the authoring guide for how to create and publish your own stack.

agentic-stacks create my-org/my-stack
# edit skills, CLAUDE.md, stack.yaml
agentic-stacks doctor --path ./my-stack
agentic-stacks publish --path ./my-stack

Development

pip install -e ".[dev,local,mcp]"
pytest -v --tb=short

License

MIT — see LICENSE.