infrahub

Infrahub AI Skills

AI skills for developing with Infrahub. Install the skills, open your Infrahub project, and start building — your AI coding assistant automatically uses the right skill for each task.

# Install all Infrahub skills into your project
npx skills add opsmill/infrahub-skills

The skills installer will walk through where you want to install the skill (project specific or global), and what agents you want to be able to use it.

To update teh skills:

# Update skills
npx skills update

Once installed, start working on an infrahub project — try "describe what this schema does" to explore, or "add a status attribute to Device" to make a change.

Note, the npx skills command uses the Vercel Skills CLI. See the Vercel Skills CLI documentation for details

What You Can Do With It

• Build a working schema from a description — describe your data model in plain terms and the Schema Creator produces valid Infrahub schema YAML with appropriate node types, attribute kinds, and relationships, without requiring manual study of the schema format first.
• Generate automation logic from a plain description — describe what you want to automate (for example, "create a BGP session for each spine-leaf pair in my fabric design") and the Generator Creator produces a working InfrahubGenerator implementation for that specific case.
• Get working configuration templates for your data model — describe the output format you need and the Transform Creator produces a transform and Jinja2 template that reads from your specific schema, rather than a generic placeholder example.
• Write Infrahub CI pipeline checks without SDK expertise — describe what a proposed change should or should not allow and the Check Creator produces a working InfrahubCheck implementation with the correct GraphQL queries and .infrahub.yml registration.
• Work through design decisions before building — pair the skills with a spec-driven development (SDD) framework to plan before you build: capture requirements, validate the design against Infrahub conventions, and break the work into discrete tasks before any code is generated.
• Query and analyze a live Infrahub instance — the Analyst skill connects to a running instance via MCP and answers operational questions directly: cross-node correlation, drift detection, blast-radius analysis, and data quality audits.
• Audit an existing Infrahub repo for best practices — the Repo Auditor checks your repository against Infrahub conventions: schema naming, relationship identifiers, .infrahub.yml registration, and structural patterns — then generates a report regarding what to fix and why.

How It Works

The skills give your AI agent domain-specific knowledge about Infrahub: schema design, data modeling, validation checks, generators, transforms, and menu customization. When you make a request inside an Infrahub project, the agent picks the appropriate skill and applies Infrahub conventions automatically — correct naming, relationship patterns, .infrahub.yml registration, and so on.

There are two ways to work, depending on the complexity of what you're building.

Direct Mode — Just Ask

For targeted changes, skip the ceremony. Describe what you want and the agent handles it using the right skill.

Examples:

• "Add a contract_start_date attribute to InfraCircuit" — the agent uses managing-schemas, applies naming conventions, and updates the schema file.
• "Create a check that validates every device has a primary IP" — the agent uses managing-checks, writes the Python class and GraphQL query, and registers it in .infrahub.yml.
• "Add a menu section for IP address management" — the agent uses managing-menus and produces the YAML with correct icon references and hierarchy.

This is the fastest path for well-scoped work: adding attributes, writing a check, populating objects, creating a transform. No planning step needed. It's also how most people start — install the skills, describe what you need, and iterate from there.

Spec-Driven Development — Plan, Then Build

For anything beyond a simple, self-contained change — designing a new schema node with relationships, building a generator chain, or standing up an entire new domain — you should use a spec-driven development (SDD) framework to plan first, then let the agent execute against the plan.

The SDD workflow forces the agent to reason before it builds. You write a natural-language spec describing what you want, the agent produces a plan validated against Infrahub skills, breaks it into discrete tasks, and then implements each one using the correct skill. This matters because a schema node with incorrect relationship cardinality or a generator missing allow_upsert will cost you debugging time later — the planning step catches those issues upfront.