| name | infrahub-managing-schemas |
| description | Creates, validates, and modifies Infrahub schema YAML files — nodes, generics, attributes, relationships, and extensions. TRIGGER when: designing data models, adding schema nodes, validating schema definitions, planning schema migrations, modeling file objects / attachments / uploads (storing PDFs, diagrams, images, certificates, documents as Infrahub objects). DO NOT TRIGGER when: populating data objects, writing checks/generators/transforms, querying live data. |
| allowed-tools | ["Read","Write","Edit","Bash"] |
| argument-hint | [namespace] [node-names...] |
| metadata | {"version":"1.2.7","author":"OpsMill"} |
Infrahub Schema Creator
Overview
Expert guidance for designing and building Infrahub
schemas. Schemas are YAML files defining nodes (concrete
types), generics (abstract base types), attributes,
relationships, and extensions.
Project Context
Existing schemas in this project:
!find . -name "*.yml" -path "*/schemas/*" -o -name "*schema*" -name "*.yml" 2>/dev/null | head -20
Infrahub config (if present):
!cat .infrahub.yml 2>/dev/null || echo "No .infrahub.yml found"
If invoked with arguments (e.g., /infrahub:managing-schemas Ipam Vlan VlanGroup),
use the first argument as the namespace and remaining arguments as node names.
When to Use
- Designing new data models or schema nodes
- Adding attributes or relationships to existing schemas
- Setting up hierarchical location trees or component/parent patterns
- Configuring display properties (human_friendly_id, display_label)
- Migrating or refactoring existing schemas
- Debugging schema validation errors
Rule Categories
| Priority | Category | Prefix | Description |
|---|
| CRITICAL | Branch-First Changes | workflow- | Load schema onto a branch, not the default branch |
| CRITICAL | Naming | naming- | Namespace, node, attribute naming |
| CRITICAL | Relationships | relationship- | IDs, peers, component/parent, on_delete |
| HIGH | Attributes | attribute- | Defaults, dropdowns, computed Jinja2, branch-agnostic, deprecated |
| HIGH | Hierarchy | hierarchy- | Hierarchical generics, parent/children |
| HIGH | Display | display- | human_friendly_id, order_weight, menu placement |
| MEDIUM | Extensions | extension- | Cross-file via extensions block, artifact targets |
| MEDIUM | Uniqueness | uniqueness- | Constraint format, __value suffix |
| MEDIUM | Migration | migration- | Add/remove attributes, state: absent |
| HIGH | Validation | validation- | Load-time string-length caps (description / label / identifier), common error messages, pre-check checklist |
Schema File Basics
---
version: "1.0"
generics:
- ...
nodes:
- ...
extensions:
nodes:
- ...
Always include the $schema comment for IDE validation.
Only version is required at the top level.
Designing for Downstream Consumers
A schema node rarely lives alone. Before finalizing it,
walk through how it will be used by other parts of the
project and add the inheritance / configuration that
those features require:
This audit is the difference between a schema that
"validates" and one that "actually works in the broader
project." Skipping it forces a schema migration once the
downstream feature is wired up — at which point the data
is already loaded.
When the task spans multiple skills (schemas + transforms,
schemas + menus, etc.), load both skills' rules together
rather than treating the boundaries as exclusive.
Design for the cheaper layer
A schema choice can remove the need for Python or
denormalized data downstream. The schema is the cheapest
place to get this right — fixing it later means a
migration on already-loaded data. Before adding a field or
node, check whether a built-in or structural feature
already covers it:
These are the schema-side counterparts to the "Before
writing Python" guidance in the checks, transforms, and
generators skills. The repo auditor flags them as advisory
cost-to-fix findings; catching them at design time avoids
both the finding and the later migration.
Workflow
Follow these steps when creating or modifying a schema:
- Gather requirements — Identify the node types,
their attributes, and how they relate to each other.
Ask about hierarchies, dropdowns, and display needs.
- Read relevant rules — Read
rules/naming-conventions.md
for naming constraints,
rules/attribute-defaults-and-types.md
for attribute kinds and defaults, and
rules/relationship-identifiers.md
for bidirectional relationship setup.
- Build the schema YAML — Start with the
$schema
comment and version: "1.0". Define generics first
(if any), then nodes. Apply naming, display, and
relationship rules from step 2.
- Audit downstream consumers — Walk the table in
"Designing for Downstream Consumers" above. If any
node will become an artifact or generator target, add
CoreArtifactTarget to its inherit_from now, per
rules/extension-artifact-target.md.
Adding it later forces a schema migration on loaded data.
- Configure display properties — Set
human_friendly_id, display_label, and
order_weight per
rules/display-human-friendly-id.md
and rules/display-order-weight.md.
- Validate and roll out on a branch — Run
infrahubctl schema check to fix errors per
validation.md and
rules/validation-common-errors.md.
Then apply the change on a dedicated branch, not the
default branch (main by convention, but it can be
renamed): infrahubctl branch create <name> →
schema check --branch <name> →
schema load --branch <name>, and merge via a proposed
change once it looks right. A schema load runs
migrations against loaded data immediately, so on a
shared server the default branch gives no preview and no
per-step undo — the branch does. See
rules/workflow-branch-first.md.
The default branch is only reasonable on a local
throwaway instance.
Production Patterns Worth Knowing
Seven recurring patterns — computed Jinja2 attributes,
cascade-vs-no-action deletes, menu visibility,
branch-agnostic identity, artifact targets, object
templates, and file objects — are documented at the top
of examples.md. Read those before
finalizing a schema; each pattern is easy to miss
when building from scratch and expensive to retrofit
after data is loaded.
Supporting References
