This example demonstrates the final step of the AgentFigureGallery loop:
agent query -> gallery display -> human like/reject/select -> export bundle -> plot code
The plot is intentionally small and synthetic so it can run from a clean clone. It uses selected embedding references as visual guidance and preserves their candidate IDs in the figure note.
Selected visual guidance from reference_bundle_example.json:
EMB-3928F88AD7EMB-FE34642DA4
python examples/generated_embedding_plot/plot_embedding.pyOutputs:
examples/generated_embedding_plot/figures/embedding_preview.pngexamples/generated_embedding_plot/figures/embedding_nature_style.pdfexamples/generated_embedding_plot/figures/embedding_nature_style.svg
- muted multi-class palette
- light grey background cells with highlighted disease outlines
- direct cell-type labels at cluster centroids
- compact frameless legend
- stable reference IDs retained in the plot annotation
In a real session, agentfiguregallery bundle --session <session_id> writes:
outputs/reference_sessions/<session_id>/export_bundle/reference_bundle.json
This example keeps a tiny reference_bundle_example.json beside the script so the final plot can be regenerated without a live browser session.
