Skip to main content

maf-sandbox-drawio

PyPI Python License

Experimental. Releases before 1.0 may change or remove APIs. Importing this package emits MafSandboxDrawioExperimentalWarning.

Create an editable diagram.drawio from model-supplied XML. The create_drawio(xml: str) tool validates draw.io cells, applies layout when needed and delivers the file through the host's OutputSink.

Requires Python 3.12–3.14.

pip install maf-sandbox-drawio

Attach the tool

from maf_sandbox_drawio import make_drawio_tools

tools = make_drawio_tools(
    router,
    agent_id="diagram-designer",
    context=context,
    sink=sink,
    image="drawio-sandbox:local",
    preserve_layout=True,
    direction="TB",
)

The host supplies the router, CallerContext and sink. The model supplies only XML; storage and layout settings belong to the host.

Use make_file_system_sink(output_directory, existing="replace") to replace the previous diagram, or a per-call sink to retain separate outputs. The result contains the sink's display reference after delivery.

Build the image from the repository root:

docker build -t drawio-sandbox:local images/drawio-sandbox

The image needs Python and Graphviz. The kind requires a POSIX backend with EXEC, FILES_IN and FILES_OUT. It uses closed network access and the host's isolation floor. Core disposes after each call by default.

For Docker, use await DockerSandboxBackend.create(config) so the backend declares POSIX. The plain constructor does not declare an OS family.

Optional image exports

Build the export runtime with docker build -t maf-drawio-export:local images/drawio-export. The image's coordinator runs as root so ACAS's root-owned file-plane directories are usable, while Xvfb and Electron run as uid/gid 10001 (#1710). An export image built before that change is refused on ACAS, so rebuild and re-import it, under a new tag. Then opt in through host configuration:

from maf_sandbox_drawio import DrawioExport, make_drawio_tools

tools = make_drawio_tools(
    router,
    agent_id="diagram-designer",
    context=context,
    sink=sink,
    image="maf-drawio-export:local",
    export=DrawioExport(formats=("png", "jpg", "svg")),
    exec_timeout_seconds=120,
)

Each selected page produces diagram-1.png, diagram-1.jpg and diagram-1.svg (and so on), alongside the editable diagram.drawio. Formats are a nonempty unique tuple of png, jpg and svg. Pages default to all pages; pages=(2,) selects the second page. The host can set scale in (0, 4], transparent=True for PNG/SVG, and jpeg_quality from 1 to 100. JPG uses an opaque background. File limits become 25 outputs, 8 MiB per file and 32 MiB total; each render is limited to 4096 pixels per axis and 16 million pixels. The execution deadline covers layout and all exports.

To export an existing file, attach make_drawio_export_tools(router, agent_id, context, sink, store, export=DrawioExport(...), image=...). Its export_drawio(file) argument must match exactly one visible store reference. The store supplies uncompressed XML text with complete geometry; the tool reads it through the caller context and does not run layout again. The stored source is not overwritten, although the returned editable copy may have normalized XML formatting.

Pass the host's file_store_provenance and optional requires_file_integrity to apply the core's recorded-read and admission policy. Resource validation does not promote source integrity.

Both export-enabled creation and stored-file export require a backend that supports call isolation. Each call receives its own sandbox, so overlapping editable/export tools can use different runtime images. Editable-only creation retains its existing conversation scope and cleanup policy.

The offline profile accepts bundled cloud icons (including img/lib/azure2/...), validated embedded PNG/JPEG/SVG images and formatting-only HTML labels. Known https://app.diagrams.net/img/lib/... references map to bundled assets without a request. External resources, missing assets, unsupported shapes/fonts, math, enabled label placeholders, links and background resources are refused. Fonts are DejaVu Sans, Serif and Sans Mono; common Arial/Helvetica, Times New Roman and Courier New requests map to those families. Other languages may require a future expanded font profile.

SVG embeds images and font data. Rich labels use foreignObject, so use a compatible browser viewer; arbitrary SVG consumers may display them differently. Raster and SVG export use the same native document renderer. Every requested output is checked before sink delivery, but the sink is not transactional: a later delivery failure can leave earlier files saved and returns an incomplete result.

Run uv run python scripts/check_drawio_exports.py --image maf-drawio-export:local --output out/drawio-exports to exercise real offline rendering. The image is Linux amd64; container-free execution is tracked separately in #1656. A backend must provide the installed runtime and enforce closed networking and process cleanup. Export support does not change the host's isolation floor.

Model input

Accepts one uncompressed mxGraphModel or an mxfile containing uncompressed pages:

<mxGraphModel>
  <root>
    <mxCell id="0"/>
    <mxCell id="1" parent="0"/>
    <mxCell id="a" value="Start" vertex="1" parent="1"/>
    <mxCell id="b" value="Finish" vertex="1" parent="1"/>
    <mxCell id="edge" edge="1" source="a" target="b" parent="1"/>
  </root>
</mxGraphModel>

Editable-only output keeps native shapes, connectors, labels, styles, IDs, metadata and layers. This mode does not render a preview or fetch external resources. XML references remain in the artifact, so editable-only validation is not content sanitization. Optional image export applies the offline resource profile described above.

Layout policy

Input geometry preserve_layout=True preserve_layout=False
Complete Keep supplied positions and sizes Apply automatic layout
Incomplete Apply automatic layout Apply automatic layout

Complete geometry needs positive width and height for each vertex. Missing x or y means zero. Overlapping vertices are accepted. Missing connector waypoints do not trigger layout; missing edge geometry receives standard relative geometry.

Automatic layout handles flat flowcharts and component graphs, including cycles and disconnected components. direction="TB" places the graph top to bottom; "LR" places it left to right.

Graphviz chooses positions and connector paths. Supplied dimensions and rotation are retained; missing dimensions use 160 by 80. Automatic layout replaces routing hints while keeping appearance styles.

Nested groups, relative ports, edge-label vertices, collapsed cells and detached edges require complete geometry with preservation enabled. They are refused when automatic layout is needed.

The tool provides no sequence, BPMN or ER-specific layout, automatic label fitting or guarantee against every overlap. Give large labels explicit dimensions. Preserved geometry keeps its values, not byte-for-byte XML formatting.

Validation and limits

Every page is checked before and after layout. Invalid IDs, parent cycles, broken endpoints, unsupported geometry and non-finite coordinates are refused. DTDs, entities and compressed pages are refused too.

Limit Value
Input 1 MiB, up to 8 pages and 1,000 cells per page
Automatic layout Up to 200 vertices and 600 edges per page
Output One file, at most 2 MiB
Execution One deadline for all pages: 60 seconds by default, at most 300
Diagnostics At most 2,048 characters

A failure on any page prevents delivery of the entire artifact. create_drawio returns separate content items: trusted completion and verdict, any fixed explanation, and separately labelled sink display text or converter diagnostics. With maf-sandbox 0.46.1 or later, a call with trusted FIDES conversation and argument labels returns trusted display text and diagnostics. Expanded untrusted arguments, an untrusted conversation or absent call evidence retain the untrusted workload claim. The verdict is created after delivery and refused for rejected XML or unsupported layout requests. Invalid tool arguments, execution failures, missing output and delivery failures remain incomplete with no verdict. The host sets confidentiality and destination policy.

Verify all four layout-policy cases against Docker from a repository checkout:

uv run python scripts/check_drawio_docker.py --image drawio-sandbox:local --output out/drawio

See the kind guide for accepted geometry and labels, and the image definition for pinned dependencies.

Metadata

Release files for maf-sandbox-drawio 0.4.1

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for maf-sandbox-drawio 0.4.1
File Size Uploaded
maf_sandbox_drawio-0.4.1.tar.gz 17.9 kB Details

Built distribution (wheel)

Table of built distributions (wheels) for maf-sandbox-drawio 0.4.1
File Interpreter ABI Platform
maf_sandbox_drawio-0.4.1-py3-none-any.whl Python 3 none any Details

Total release size: 38.0 kB

Release files / maf_sandbox_drawio-0.4.1.tar.gz

Download URL maf_sandbox_drawio-0.4.1.tar.gz
Size 17.9 kB
Tags Source
SHA-256 checksum
How to use checksums
41de907eed89306d8761769f50b77d84cde8b9443df25329fcc3b7963e8558b3
BLAKE2b-256 checksum
How to use checksums
51d33c24d62f8466b512020fbb859404f0b9fb3f9108e44729423cb45c3d931b
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Oct 5, 2026.

Transparency log

Release files / maf_sandbox_drawio-0.4.1-py3-none-any.whl

Download URL maf_sandbox_drawio-0.4.1-py3-none-any.whl
Size 20.1 kB
Tags Python 3
SHA-256 checksum
How to use checksums
5ac4dea109d062a298781c2c715e710609a79371d9b87144e08ecabf7110370f
BLAKE2b-256 checksum
How to use checksums
7851d548b9d4109f170dbfdb7b772a508812153907d28cb1b032835d282ef698
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Oct 5, 2026.

Transparency log

Release history Release notifications | RSS feed

0.5.0

2 release files

0.4.4

2 release files

0.4.3

2 release files

0.4.2

2 release files

This release

0.4.1 This release

2 release files

0.4.0

2 release files

0.3.4

2 release files

0.3.3

2 release files

0.3.2

2 release files

0.3.1

2 release files

0.3.0

2 release files

0.2.0

2 release files

0.1.0

2 release files

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page