Infrahub Transform Creator

Overview

Expert guidance for creating Infrahub transforms. Transforms convert Infrahub data into different formats -- JSON, text, CSV, device configs, or any text-based output -- using Python classes or Jinja2 templates.

Project Context

Infrahub config: !cat .infrahub.yml 2>/dev/null || echo "No .infrahub.yml found"

Existing transforms: !find . -name "*.py" -path "*/transforms/*" -o -name "*.j2" -path "*/templates/*" 2>/dev/null | head -20

When to Use

• Building data transformations (Infrahub data -> another format)
• Generating device configurations from infrastructure data
• Creating CSV reports, cable matrices, or inventory exports
• Rendering Jinja2 templates with query data
• Combining Python logic with Jinja2 rendering
• Connecting transforms to artifacts for automated output

Rule Categories

Priority	Category	Prefix	Description
CRITICAL	Types	types-	Python vs Jinja2 choice
CRITICAL	Python	python-	InfrahubTransform class
CRITICAL	Jinja2	jinja2-	Template syntax, filters
HIGH	Hybrid	hybrid-	Python + Jinja2 combined
HIGH	Artifacts	artifacts-	Output files, targets
HIGH	API Ref	api-	Class attrs, methods
MEDIUM	Patterns	patterns-	Utilities, CSV, shared
HIGH	Testing	testing-	Resources Testing Framework, transform/render commands

Schema Features This Skill Depends On

A transform reads schema-shaped data and produces a file. Misalignment between the transform and the schema fails late — at artifact-render time, when someone is waiting for the output.

If the transform...	The schema (or .infrahub.yml) must...	See
Will feed an artifact_definitions entry	The target node must inherit_from: CoreArtifactTarget so the artifact pipeline can attach to it	../infrahub-managing-schemas/rules/extension-artifact-target.md
Reads attributes from a node	Define those attributes with their full __value access path in GraphQL — silent empty strings come from accessing the node, not the value	../infrahub-managing-schemas/rules/attribute-defaults-and-types.md
Picks a template per device by platform/role	The schema must expose that platform/role as a real attribute or relationship — string-matching on display_label is brittle	../infrahub-managing-schemas/rules/display-human-friendly-id.md
Is referenced from artifact_definitions.transformation	The transform's registered name must match the transformation: field exactly — mismatch produces "transformation not found" at render time	rules/artifacts-definitions.md
Uses Jinja2 (not Python)	Register under jinja2_transforms with a top-level query: field — python_transforms binds query on the class, the two keys are not interchangeable	rules/api-reference.md
Is a Python transform	Carry a watch: block naming every first-party module it imports — sibling modules included, since imports are never followed. Without the key, Infrahub cannot trust the dependency list and re-renders the artifacts on every commit	rules/artifacts-watch-dependencies.md
Reads a data file at runtime, or includes a template by variable name	Name those paths in watch.files — detection cannot see a path that exists only as a string, so the artifacts go stale on a change with no error raised	rules/artifacts-watch-dependencies.md

Before writing Python

If the transform body is string formatting — f-strings, concatenation, conditional sections — Jinja2 expresses the same output in fewer lines, renders directly in the proposed-change UI, and lives under jinja2_transforms in .infrahub.yml instead of python_transforms. Walk this ladder before reaching for InfrahubTransform:

Signal	Cheaper layer	See rule
Transform body is return f"..." or "\n".join([...]) built from query results	Jinja2 template file	yagni-python-transform-that-could-be-jinja2
Transform copies query data verbatim without computation	The GraphQL query alone — no transform needed	Ladder step 1 (drop the requirement); judgment call, no rule
Conditionals are if x: out += ...; else: out += ... and nothing else	Jinja2 {% if %} blocks	yagni-python-transform-that-could-be-jinja2

Use Python when the transform parses, computes, or reshapes — IP/subnet math, hashing, ordered aggregation, structural JSON re-shaping. See rules/python-transform.md for the legitimate cases.

When the transform reads objects through the SDK, type those calls with generated protocol classes rather than string kinds — client.filters(NetworkLink, ...), not kind="NetworkLink" — so schema drift fails type-check instead of at runtime. See protocols-adopt-typed-kinds.

Transform Basics

Two types of transforms:

Type	Output	Entry Point
Python	Whatever the artifact's content_type asks for: a dict only under application/json or application/yaml, a str for the other six	InfrahubTransform.transform()
Jinja2	Text	.j2 template file

from infrahub_sdk.transforms import InfrahubTransform

class MyTransform(InfrahubTransform):
    query = "my_query"

    # Returns a dict, so the artifact definition has to declare
    # content_type: application/json (or application/yaml). Under any
    # other content type this dict is stored as str(dict), with no
    # error. See rules/artifacts-definitions.md.
    async def transform(self, data: dict) -> dict:
        device = data["DcimDevice"]["edges"][0]["node"]
        return {"hostname": device["name"]["value"]}

Workflow

Follow these steps when creating a transform:

1. Choose the transform type — Python for JSON/dict or complex logic, Jinja2 for text templates, hybrid for both. Read rules/types-overview.md.
2. Write the GraphQL query — Create a .gql file that fetches the data to transform. Read ../infrahub-common/graphql-queries.md for query patterns.
3. Implement the transform — For Python, inherit from InfrahubTransform and implement transform(). Read rules/python-transform.md. For Jinja2, create a .j2 template. Read rules/jinja2-template.md. For hybrid, read rules/hybrid-python-jinja2.md.
4. Connect to artifacts — If the transform output should be stored as a file, configure artifact definitions. See rules/artifacts-definitions.md.
5. Register in .infrahub.yml — Add under python_transforms or jinja2_transforms. See rules/api-reference.md. Then declare the transform's dependencies with watch.files: read the entry point's imports and runtime file reads, and name every first-party path they resolve to. A Python transform should always carry the key — files: [] when it genuinely has no dependency beyond its own file. Read rules/artifacts-watch-dependencies.md, which also covers reviewing an existing watch block for entries that are missing, stale, or wrong.
6. Add tests — Create YAML-driven test definitions (smoke, unit, integration) alongside the transform so it is validated automatically in the proposed change pipeline. Read rules/testing-resource-framework.md.
7. Test locally — Run infrahubctl transform or infrahubctl render to validate. See rules/testing-commands.md.

Supporting References

• rules/artifacts-definitions.md -- The eight content_type values and how each serialises the transform's return value. Authority on which return type a content type accepts
• reference.md -- Class API, lifecycle, .infrahub.yml registration shapes, filter env overview
• examples.md -- Complete transform patterns (Python, Jinja2, hybrid, CSV)
• ../infrahub-common/graphql-queries.md -- GraphQL query writing reference
• infrahub-yml-reference.md -- .infrahub.yml project configuration
• ../infrahub-common/rules/ -- Shared rules (git integration, caching) across all skills
• ../infrahub-common/rules/workflow-information-priority.md -- Skill content first; how to consult docs.infrahub.app on a genuine gap (e.g. deleting nodes)
• ../infrahub-managing-schemas/SKILL.md -- Schema definitions transforms work with
• rules/ -- Individual rules by category