cvpr-figurelisted

# Conference figure making — router The engine is deterministic: you write **what goes where in what order**, and `scripts/cvprfig` computes geometry in final rendered points. Do not hand-write SVG coordinates and do not draw with an image model — both produce the generic look this skill exists to prevent. Read fragments from disk as directed below. Do not work from memory of this file. ## 0. Is this the right skill? | Request | Route | |---|---| | architecture / pipeline / framework / teaser / module / attention diagram | continue here | | bar chart, curve, heatmap, scatter, ablation plot, radar | not this skill — matplotlib/seaborn | | a graphical abstract for a *biomedical* journal | not this skill | | "make my figure Visio-editable" | continue here, jump to step 5 | | "draw the architecture from this code / config" | continue here, but start at step 3b | ## 1. Always load 1. [static/core/stance.md](static/core/stance.md) — the non-negotiables and the list of tells this skill exists to suppress. 2. [static/core/contract.md](static/core/contract.md) — the seven questions to answer before drawing. Then read [manifest.yaml](manifest.yaml) to see which further fragment the current job needs. ## 2. Fill in the contract — do not skip Answer, in your reply to the user, before writing any spec: 1. the one-sentence claim the figure must make; 2. the archetype; 3. which element is the contribution; 4. the semantic colour map; 5. venue and column width; 6. which real