linkml-mermaid

Render Mermaid diagrams and GFM tables from LinkML instance data, driven by annotations in the schema itself.

Why this exists

LinkML already generates diagrams — seven generators, in fact, including gen-erdiagram and gen-mermaid-class-diagram. Every one of them renders the schema: classes, slots, ranges, inheritance. That is the right answer when the question is "what does my model look like?".

It is the wrong answer when the question is "what does my data look like?".

A LinkML schema that models a workflow says a Transition has a from_state and a to_state. Only the instance data says Draft goes to InReview when a Reviewer approves it. A class diagram of that schema shows three boxes — Workflow, State, Transition — and tells a reader nothing about the workflow. The diagram people actually want is a state machine, and it can only be drawn from the instances.

So teams write it by hand. The schema is the source of truth, the diagram in the README is a copy of it, and the two drift apart on the first change that nobody remembers to mirror. The usual fix is a one-off script with slot names hardcoded into f-strings — which works until the schema is renamed, and which has to be rewritten from scratch for the next project.

This package closes that gap. You annotate the schema once to say which slot means "source state" and which means "guard", and the diagram is generated from the data every time it is needed.

This is a known gap, not an invented one

linkml/linkml#915 — "Add dumpers for mermaid, markdown", opened by LinkML's lead maintainer in August 2022 — is still open.

To be straight about it: that issue asks for instance-data rendering in general. We have found no issue specifically requesting state diagrams, so this package is not filling a loudly-articulated demand for that shape. It is an answer to the general request, for the case we needed solved.

Prior art

linkml/linkml-renderer is the only existing tool that renders LinkML instance data to Mermaid, and it deserves a look before you reach for this one. It walks an object tree and emits Mermaid. The differences that matter:

If you want a generic object graph, use linkml-renderer. If you are modelling a state machine and want it drawn as one — with guards on the arrows and a table of states beside it — that is what this package does.

LinkML's own SQLTableGenerator already reads unprefixed annotations such as primary_key, so the mermaid_role convention used here follows established LinkML practice rather than inventing a new mechanism.

Install

pip install linkml-mermaid

Use it from the command line

linkml-mermaid diagram -s schema.yaml -d data.yaml -c TrafficWorkflow --direction LR

stateDiagram-v2
    direction LR
    [*] --> RED

    RED : Red
    YELLOW : Yellow
    GREEN : Green
    RED --> GREEN : Go [Timer]<br>Timer expired
    GREEN --> YELLOW : Caution [Timer]
    YELLOW --> RED : Stop [Timer]

linkml-mermaid table -s schema.yaml -d data.yaml -c TrafficState -r states

| State | Description | Color |
| --- | --- | --- |
| **Red** | RED | #d32f2f |
| **Yellow** | YELLOW | #fbc02d |
| **Green** | GREEN | #388e3c |

-o FILE writes to a file instead of standard output, which is what you want in a pre-commit hook or a docs build. Run linkml-mermaid diagram --help for the remaining options.

Annotate your schema

mermaid_role declares what Mermaid concept a slot carries. The reader discovers the rest.

classes:
  Workflow:
    annotations:
      mermaid_diagram: stateDiagram-v2
    attributes:
      states:
        range: State
        multivalued: true
        annotations: { mermaid_role: states }
      transitions:
        range: Transition
        multivalued: true
        annotations: { mermaid_role: transitions }
      guards:
        range: Guard
        multivalued: true
        annotations: { mermaid_role: guard_list }

  State:
    attributes:
      id:
        range: MyStateEnum
        annotations: { mermaid_role: state_id }
      display_name:
        annotations: { mermaid_role: label }
      is_initial:
        range: boolean
        annotations: { mermaid_role: initial }
      is_terminal:
        range: boolean
        annotations: { mermaid_role: terminal }

  Transition:
    attributes:
      name:
        annotations: { mermaid_role: transition_label }
      from_state:
        range: MyStateEnum
        annotations: { mermaid_role: from_state }
      to_state:
        range: MyStateEnum
        annotations: { mermaid_role: to_state }
      actor:
        annotations: { mermaid_role: actor }
      guards:
        range: MyGuardEnum
        multivalued: true
        annotations: { mermaid_role: guards }

  Guard:
    attributes:
      id:
        range: MyGuardEnum
        annotations: { mermaid_role: guard_id }
      display_name:
        annotations: { mermaid_role: guard_label }

Vocabulary

All 14 mermaid_role values are defined in vocab/mermaid_annotations.yaml — a real LinkML schema, so the vocabulary is itself machine-readable — and mirrored as Python frozensets in linkml_mermaid.vocab.

Two further keys exist: mermaid_diagram on the container class (only stateDiagram-v2 is accepted), and table_column on any slot, whose value is Header or Header|format.

Use it from Python

import yaml

from linkml_mermaid import (
    ColumnDef,
    MarkdownTableRenderer,
    SchemaReader,
    StateDiagramRenderer,
    build_state_descriptions,
    build_table_rows,
    map_states,
    map_transitions,
)

reader = SchemaReader("schema.yaml")
config = reader.discover_state_diagram_config("Workflow")

with open("data.yaml") as fh:
    data = yaml.safe_load(fh)

diagram = StateDiagramRenderer(direction="LR").render(
    map_states(data["states"], config),
    map_transitions(data["transitions"], data["states"], data.get("guards", []), config),
)
print(diagram)

rows = build_table_rows(
    data["states"],
    {"state": "display_name", "description": "id"},
    lookups={"description": build_state_descriptions(reader, config.state_enum_name)},
)
print(
    MarkdownTableRenderer().render(
        [
            ColumnDef("State", "state", format="bold"),
            ColumnDef("Description", "description", align="left"),
        ],
        rows,
    )
)

Instances may be dicts, dataclasses or Pydantic models — the mapping layer reads all three.

Flowcharts, without LinkML

The renderers are a standalone layer with no LinkML dependency. Build the dataclasses yourself and render them.

from linkml_mermaid import FlowchartEdge, FlowchartNode, FlowchartRenderer, FlowchartSubgraph

nodes = [
    FlowchartNode("A", "Start", shape="stadium"),
    FlowchartNode("B", "Process"),
    FlowchartNode("C", "Decision", shape="diamond"),
    FlowchartNode("D", "End", shape="round"),
]
edges = [
    FlowchartEdge("A", "B"),
    FlowchartEdge("B", "C"),
    FlowchartEdge("C", "D", label="Yes"),
    FlowchartEdge("C", "B", label="No", style="dotted"),
]
subgraphs = [FlowchartSubgraph("main", "Main Flow", ["A", "B", "C", "D"])]

print(FlowchartRenderer(direction="TD").render(nodes, edges, subgraphs))

Shapes: rect ["…"], round ("…"), stadium (["…"]), diamond {"…"}, hexagon {{"…"}}, subroutine [["…"]], circle (("…"))

Edge styles: solid -->, dotted -.->, thick ==>

Escaping is not optional

A state called Needs "sign-off" or a table cell containing a | b will produce broken output if it is interpolated into a template. Every label, cell and title this package emits goes through the escaping layer, which encodes " as #quot; and # as #35; for Mermaid (per the flowchart spec's entity-code rule) and escapes pipes — including backslash runs preceding them — for GFM.

Those helpers are public, so code that composes its own output can reuse them:

from linkml_mermaid import escape_mermaid_text, escape_state_label, escape_table_cell

Architecture

Layer 1 — Pure renderers (no LinkML dependency)
  types.py            Frozen dataclasses: MermaidState, MermaidTransition,
                      FlowchartNode, FlowchartEdge, ColumnDef, …
  escaping.py         The escape rules every renderer shares
  state_diagram.py    StateDiagramRenderer → Mermaid stateDiagram-v2
  flowchart.py        FlowchartRenderer → Mermaid flowchart
  markdown_table.py   MarkdownTableRenderer → GFM §4.10 pipe tables

Layer 2 — Config-driven LinkML mapping
  reader.py           SchemaReader (SchemaView façade + annotation discovery)
  mapping.py          map_states, map_transitions, build_table_rows, …

Layer 3 — Command-line interface
  cli.py              linkml-mermaid diagram | table

Shipped vocabulary
  vocab/              mermaid_annotations.yaml and its Python mirror

Standards compliance

Claims about compliance are only worth the evidence behind them, so the evidence is in the repository. docs/standards/ pins each specification by SHA-256, maps every implemented clause to the code and the test that prove it, and records the clauses that are deliberately not implemented. Start with docs/standards/coverage.md. scripts/check_standards.py enforces the whole record in CI and will fail if an upstream specification changes under a quoted citation.

• Mermaid stateDiagram-v2 — https://mermaid.js.org/syntax/stateDiagram.html
• Mermaid flowchart — https://mermaid.js.org/syntax/flowchart.html
• GFM §4.10 Tables — https://github.github.com/gfm/#tables-extension-
• LinkML annotations — https://linkml.io/linkml/schemas/annotations.html

What this package does not do

• Composite states, choice, fork/join, concurrency. Branching is expressed as guarded transitions. Nesting is not modelled.
• Styling. No classDef is emitted; append your own to the output.
• Images. This emits Mermaid text. Rendering it to SVG or PNG is Mermaid's job, or mermaid-py's.
• Annotation-driven flowcharts. mermaid_diagram accepts stateDiagram-v2 only; flowcharts are built through the Python API.
• Open links. Every edge is directed; A --- B is not emitted.
• Parser-level validation. Output is tested against grammar patterns derived from the specifications and against adversarial input, but no JavaScript Mermaid parser runs in CI.

The full list, with reasons, is in docs/standards/README.md.

Development

uv sync --group dev
uv run ruff check .
uv run mypy --strict src
uv run pytest --cov
uv run python scripts/check_standards.py

See CONTRIBUTING.md.

License

MIT — see LICENSE.