What problem does it solve? After a computational-biology stage finishes, someone must visually inspect what it produced — checking whether a cluster is real, whether a threshold holds, or how conditions compare — and summary tables alone cannot answer these questions. This Skill provides a disciplined, read-only notebook pattern for that inspection so exploration never corrupts authoritative pipeline outputs. ## Core Features & Use Cases - Two exploration flavors: a live-kernel Python notebook with jscatter linked panels for brushing and lassoing cells, or a rendered Quarto/R notebook that re-plots stage results into committed, GitHub-shareable markdown and HTML. - OOM-safe widget lifecycle: enforces a grid()/close_panels() contract so re-running jscatter panels never leaks ipywidgets and kills the kernel, plus an export-first rule that loads compact parquet tables instead of multi-GB .h5ad checkpoints. - Durable conclusions: selections are saved as barcode-indexed CSVs and the analytical reading is recorded in a stage-keyed topic note, keeping evidence reproducible across sessions. - Use Case: After a QC stage flags a high-mitochondrial FOXP3 population, scaffold a live jscatter explorer to lasso the pocket, characterize it against the full dataset in pandas, save the barcodes, and write the verdict into a durable note. ## Quick Start Ask the user what they want to see and whether they prefer a live Python kernel or a rendered Quarto notebook, then scaffold the matching template from the bundled assets.
