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 — ask the agent to train you on any domain and it builds an interactive curriculum from the stack's skills.

How It Works

# Start a project and pull a stack
agentic-stacks init my-cluster
cd my-cluster
agentic-stacks pull kubernetes-talos

# Now talk to the agent — it knows Kubernetes on Talos
# "I need a 5-node HA cluster with Cilium CNI and
#  local-path storage on these 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.

Learn from your stacks

Every project includes common-skills — a shared stack with training, guided walkthroughs, orientation, and feedback capture. Ask the agent to teach you the domain and it builds a curriculum from the stack's skills.

# In a project with stacks pulled:
> train me on this stack
> train me on RAID management
> quiz me
> what should I learn next?

The agent assesses what you already know, sequences topics from foundational to advanced, and adapts as you go. Stacks teach agents and humans.

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

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

Install

pipx install agentic-stacks

CLI

# Start a new project
agentic-stacks init my-project
cd my-project

# Pull stacks into .stacks/
agentic-stacks pull kubernetes-talos    # pull a stack
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:

kubernetes-talos/
├── CLAUDE.md               # Agent entry point — the expertise guide
├── stack.yaml              # Manifest — identity, skills, metadata
└── skills/                 # Markdown knowledge — teaches the agent
    ├── deploy/             # Bootstrap, Networking, Storage
    ├── foundation/         # Concepts, Infrastructure, Machine Config
    ├── operations/         # Backup, Certs, Health Check, Scaling, Upgrades
    ├── platform/           # GitOps, Ingress, Observability, Security
    └── reference/          # Compatibility, Known Issues, Decision Guides

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-cluster/
├── .stacks/                # pulled stack repos (gitignored)
│   ├── kubernetes-talos/   # platform expertise
│   └── hardware-dell/      # hardware expertise
├── CLAUDE.md               # points agent to .stacks/*/CLAUDE.md
├── stacks.lock             # pinned stack references
├── controlplane.yaml       # Talos machine config (agent created this)
├── worker.yaml             # worker node config (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 kubernetes-talos           # → github.com/agentic-stacks/kubernetes-talos
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.