Audit and fix one paper's interactive simulation against a strict UI/UX and correctness rubric. Use when reviewing or repairing a visualization in components/visualizations/<slug>.tsx — wasted space (a small scroll box stranded inside a large reserved box), hidden scroll with no affordance, cramped padding, broken or meaningless color, dials that change nothing or are unlabeled, too many knobs, and logic that doesn't actually respond to the controls. Covers the render context, a control-discipline decision procedure, the full rubric, the per-sim work procedure, the shared-file collision rule, the gates, and a final scoring checklist.
Turn a research paper PDF into a finished entry on this site. Use when the user drops a PDF into papers/ and runs /add-paper papers/<file>.pdf, or @-mentions this skill plus a PDF on a fresh session. Runs the whole pipeline end to end — read the paper, think it through, break it down, write image specs, generate 2K diagrams, write a layered explainer, build a bespoke play/step visualization, register it, and build the project — so a new searchable entry appears in the sidebar with a working page.
Reason through a hard problem and then write the result clearly. Use whenever you need to think before producing prose — analyzing a paper, drafting an explainer, a doc, or a Slack message. Two halves: THINK (seven reasoning lenses, validation, incentives, second-order effects) and WRITE (plain conversational voice, banned words, sentence discipline, layered teaching). Read the precedence note first when writing a research explainer page, because the analyst template governs structure while these voice rules govern the sentences.
Analyze, summarize, explain, visualize, or teach a research paper. Use when given a paper (or a topic from a paper) and asked for a headline, an executive summary, a structured extraction, an audience-calibrated explanation (ELI5 up to peer researcher), a figure or interactive visualization plan, a teaching breakdown, or a compressed practitioner version. Pairs with clear-thinking-writing (how to reason and how to write) and is driven by add-paper (the end-to-end pipeline).
Design and generate the figures for a paper's explainer — beautiful, minimalist 2K images that carry both the concept and the everyday analogy, with no more than 3 to 4 elements each. Use from add-paper, or directly, whenever creating diagrams for a doc. Covers the one-message rule, the element budget, fusing the mechanism with the explainer's running analogy, the house style (an inked pixel-art field-journal look on aged parchment, a tight earthy colorblind-safe palette, short black serif labels), prompt craft for the Gemini image tool, aspect ratios, descriptive alt text, the generate command, and the look-at-it-then-regenerate quality loop.
Design and build a genuinely interactive, perturbable simulation of a paper's core mechanism — a live model the reader pokes, not a slideshow. Use from add-paper (or directly) whenever a paper's idea is a dynamical system, protocol, algorithm, scheduler, or multi-agent process where the interesting thing is what happens when you change inputs, inject faults, or intervene mid-run. Covers the think-first enumeration, the model/view split, deterministic seeded time-stepping, discrete-event and physical engines, the intervention catalog (kill a node, drop a message, partition the network, perturb an entity), observability (inspect entities, event log, scrub and rewind), the controls, and the tests. The bar is that the reader can cause a different, rule-consistent outcome by acting.