Skip to content

Agent skill that turns natural language, code, Terraform/K8s, SQL, OpenAPI, AsyncAPI, Protobuf and GraphQL sources into editable, tested draw.io architecture diagrams: incremental sync, multi-view projection, drift diff, CI architecture tests, whiteboard derasterize, interactive HTML/PPTX/Mermaid exports.

1.8 KB
Raw Blame

Architecture Studio 3.0 dogfood report

Date: 2026-09-01
Target: skills/drawio-skill/scripts/ in this repository

Workflow exercised

python3 skills/drawio-skill/scripts/diagramctl.py build \
  skills/drawio-skill/scripts --from python --group \
  --ir-output /tmp/drawio-skill.ir.json -o /tmp/drawio-skill.drawio
python3 skills/drawio-skill/scripts/diagramctl.py inspect /tmp/drawio-skill.ir.json
python3 skills/drawio-skill/scripts/diagramctl.py review /tmp/drawio-skill.ir.json --format json
python3 skills/drawio-skill/scripts/diagramctl.py views /tmp/drawio-skill.ir.json \
  --views executive,system,deployment,dataflow,security -o /tmp/drawio-skill.views.drawio
python3 skills/drawio-skill/scripts/diagramctl.py story /tmp/drawio-skill.ir.json \
  -o /tmp/drawio-skill.story.html

Observed results

  • Import: 41 nodes and 1 intra-project edge.
  • Structural validation: zero errors, warnings, edge-through-node routes, crossings, or overlaps for both the system diagram and five-view diagram.
  • Views: Executive selected 12 nodes; the other four views retained all 41.
  • Story: generated offline with 41 accessible components.
  • Review: 41 owner warnings and no errors.

What the run taught us

The low edge count is plausible for this toolbox: most scripts are independent commands and only diagramctl.py imports diagram_ir.py. The importer therefore reflected the code, but assigning every module the semantic kind service made the ownership contract noisy. Likewise, deployment/data-flow/security views correctly fell back to the complete graph, but the result did not add information because code importers currently emit no runtime, boundary, or data metadata.

These are modeling/profile gaps, not layout failures. They feed the v3.1 roadmap rather than being hidden by example-specific rules.