← ClaudeAtlas

top-cs-figurelisted

Create, revise, audit, render, and export publication-ready computer-science paper figures using a Python-first, corpus-calibrated workflow. Use for manuscript figures, multi-panel experimental charts, comparison plots, scaling curves, heatmaps, embeddings, network diagrams, method schematics, executable YAML/CSV figure render specs, figure-brief handoffs, caption/callout alignment, SVG/PDF/PNG/TIFF export bundles, and visual QA for WWW, ICLR, ICML, NeurIPS, CVPR, ACL, or generic top-CS submissions. Do not use for interactive dashboards, Plotly/Altair/web apps, Illustrator/Figma-first layout, AI-generated graphical abstracts, or result-table writing.
zhiming33416/top-cs-paper-skills · ★ 1 · AI & Automation · score 74
Install: claude install-skill zhiming33416/top-cs-paper-skills
# Top CS Figure Use the skill's static/dynamic layers; do not reconstruct visual grammar or export rules from memory. ## Route the request 1. Read `manifest.yaml` and every path under `always_load`. 2. Detect `venue`, `paper_type`, and `visual_family`; resolve `figure_task`, `data_state`, `output_target`, and `figure_handoff` as runtime parameters. 3. State the route, data boundary, and output target in one short line. 4. Load only the files mapped to the selected axis values. 5. Open `references/` only when an `on_demand` condition applies, especially corpus visual style, statistics, archetypes, accessibility, provenance, revision audit, or spec rendering for figure production tasks. 6. Use `generic` for unsupported venue/year combinations and tell the user to verify current official formatting rules. ## Execute 1. Prefer an existing top-cs figure brief or render spec when provided; otherwise build a compact figure contract before plotting. Prefer the shared `figure-brief` contract when present. Preserve `figure_id`, `claim_ids`, research question, evidence status, panel jobs, caption draft, and manuscript callout. 2. If no brief exists, create a minimal figure contract before code: claim, data source, panel map, visual family, metric direction, final size, export formats, and unresolved inputs. 3. Treat data availability as binding. Use supplied data for results; when data are missing, return missing-inputs or a clearly labeled layout mockup without invented values. 4.