If you're analyzing single-cell RNA-seq and need to figure out which direction cells are differentiating without time-course data, this is the standard tool. It uses the ratio of unspliced to spliced mRNA to infer velocity vectors, letting you overlay directional arrows on your UMAP plots. The dynamical mode is more accurate than stochastic but takes 10-30 minutes for 10K cells. You'll need preprocessed counts from STARsolo or velocyto with both spliced and unspliced layers. Works seamlessly with Scanpy AnnData objects and gives you latent time, driver genes, and PAGA-compatible trajectory graphs. The phase portraits are genuinely useful for validating that your velocity estimates make biological sense.
npx skills add https://github.com/k-dense-ai/scientific-agent-skills --skill scvelo